The research, conducted by Cristian Vasilică Secu of Alexandru Ioan Cuza University of Iaşi and Radu Gabriel Pîrnău of the Romanian Academy’s Iaşi Branch, addresses a stubborn methodological problem in pedodiversity studies. Diversity indices applied to soils are notoriously sensitive to the classification system used and to the quality and resolution of available soil maps. Comparing pedodiversity between regions therefore requires that all analyzed territories share the same cartographic scale and taxonomic framework—conditions that are rarely met in practice. The Romanian team’s solution was to work within a single national classification system, the Romanian System of Soil Taxonomy, and to maintain a consistent map scale of 1:750,000 throughout the analysis.

The study area itself is a natural laboratory for soil diversity. Romania’s macrorelief is organized around the Carpathian Mountains, which occupy the central part of the country and are surrounded by the Subcarpathian hills, plateaus, and peripheral plains. Surface lithology varies dramatically across these units: mountains feature a diverse assemblage of metamorphic, igneous, and sedimentary rocks; the Subcarpathians alternate depressions and hills; lowland plains are dominated by loess and loess-like deposits; and limestone and sandstone prevail in plateau regions. The climate is temperate continental, with mean annual temperature around 9.5°C and precipitation averaging 700 mm per year, though the southeastern Dobrogea region receives less than 400 mm annually while high mountains exceed 1,000 mm.

The spatial distribution of Romanian soils follows a distinct concentric zonation around the Carpathians. Mountain regions are characterized by acidic soils—Cambisols, Podzols, and Umbrisols—with Andosols appearing locally in volcanic areas of the Eastern Carpathians. The surrounding hills and plateaus are dominated by Luvisols and Phaeozems, which gradually transition into an extensive belt of Chernozems in the peripheral lowlands. Fluvisols accompany major river valleys, often co-occurring with Gleysols and saline soils such as Solonchaks and Solonetz. Notably, the relatively similar areal proportions of Cambisols, Luvisols, and Chernozems mirror the spatial distribution of the country’s main landforms.

Methodologically, the researchers began with the SIGSTAR-200 vector database, derived from the Soil Map of Romania at 1:200,000, which contains 1,878 polygons at the soil type–subtype level. They generalized this database to 1:750,000 and then simplified the type–subtype vector to soil types alone, reducing the polygon count to 1,232. The aggregation process dissolved insular subtype polygons into their dominant encompassing polygons—for example, Cambic and Luvic Chernozem were merged as Chernozem. Crucially, the researchers preserved the surface area of each taxonomic unit across both vectors to prevent aggregation errors, and they classified taxa into two groups: variable taxa, whose polygon numbers changed during simplification, and constant taxa, which remained unchanged and typically correspond to rare or endemic soils.

The entropy calculations then proceeded along three parallel tracks. Entropy was computed separately for variable taxa extracted from the soil type vector (Y_A) and the soil type–subtype vector (Y_B), and for constant taxa (Y_C), which are common to both. The joint entropy of simultaneous distribution captured the probability that polygons for soil type–subtypes share the same spatial distribution as soil types. Finally, the connectivity index R was adapted from a method originally applied to soils and land use, approaching 1 when polygon numbers for variable taxa do not differ substantially between the two taxonomic levels and tending toward 0 when differences are large. Statistical validation included the Wilcoxon Signed-Rank Test, the Shapiro–Wilk test for normality, and rarefaction analysis in PAST 5.0 to confirm that the number of taxa estimated from polygon counts matched the values considered in the study.

The results produced a hierarchy that differs markedly from the conventional geographic regions framework. When ranked by the connectivity index, the landform units order as follows: plains (0.17) < plateaus (0.23) < Carpathians (0.27) < Subcarpathians (0.28). This sequence diverges from the geographic regions hierarchy, suggesting that connectivity between taxonomic levels is not simply a function of area or taxonomic richness. Weak connectivity—below 0.25—characterized the Southern and Eastern Carpathians, the Western Plain, and the Transylvanian Depression, indicating that numerous subtypes participate in the formation of the soil cover in these regions or that soil complexes are present, particularly in high mountain areas above 2,000 meters where glacial landforms promote Podzols and Umbrisols.

In stark contrast, the Dobrogea Plateau in southeastern Romania—the country’s most arid region, receiving less than 500 mm of precipitation annually with extreme temperatures—recorded the highest connectivity index at 0.72, followed by transitional regions such as the Getic and Moldavian Subcarpathians. The researchers attribute this pattern to the low taxonomic variability of arid regions: fewer variable taxa mean that polygon counts show minimal differences between the type–subtype and type levels, producing a relatively homogeneous structure. Interestingly, the number of variable taxa in a region does not directly determine the connectivity index value; the Getic Subcarpathians and the Danube Delta both contain three variable taxa, yet their joint entropy values differ substantially (0.48 versus 0.59), which substantially affects the connectivity index.

Bivariate correlation analysis revealed that joint entropy showed a very strong association with the connectivity index (0.98), while the diversity index of variable taxa exerted only a weak influence on both joint entropy (0.39) and the connectivity index (0.46). Constant taxa formed a separate group with a weak relationship to the diversity index of variable taxa (0.22). These findings confirm observations from the literature that probability measures alone are not always sufficient to explain the complexity of spatial structures and must be complemented by indices that capture interactions and informational redundancy. The connectivity index emerges as a valuable tool not only for describing pedological diversity but also for understanding the structural stability of spatial relationships in regions with contrasting characteristics.

The study also highlights complex interactions between landforms and parent materials. While mountain regions exhibit higher diversity due to morphological variability—in the Eastern Carpathians, volcanic and sedimentary rocks support Andosols, Eutric Cambisols, and Luvisols, whereas crystalline schists and altitudes above 2,000 meters in the Southern Carpathians promote Podzols and Umbrisols—in plains, the control exerted by parent material becomes the determining factor in structuring pedodiversity. The Romanian Plain shows a higher connectivity index (0.38) than the Western Plain (0.27) despite covering a larger area, a mismatch explained by the uniformity of aeolian loess deposits and sandy loess in the eastern plain, where the lithology associated with the highest degree of aridity limits the evolution of Chernozems toward a greater diversity of subtypes. The researchers caution that describing a region using a single diversity index value limits the ability to capture the spatial complexity of soil landscapes—a broader conceptual issue, since pedodiversity has been regarded by some authors as a controversial concept due to methodological constraints and the difficulty of simultaneously integrating pedogenetic factors, spatial variability, and taxonomic classifications.

Subject of Research: Entropy-based pedodiversity and connectivity index analysis ranking Romania's geographic regions and landform units

Article Title: Ranking Territorial Units Using Entropy-Based Pedodiversity

Article References: Secu, C. V., & Pîrnău, R. G. (2026). Ranking Territorial Units Using Entropy-Based Pedodiversity. Spanish Journal of Soil Science, 16, Article 15664. https://doi.org/10.3389/sjss.2026.15664

Image Credits: AI Generated

DOI: 10.3389/sjss.2026.15664

Keywords: pedodiversity, connectivity index, joint entropy, Shannon entropy, soil taxonomy, GIS, Romania, Carpathians, soil maps, landform units, geographic regions, information theory

The starting point of the review is a mismatch between two research traditions. Cognitive load theory, developed since the 1980s, describes how working memory limits shape learning and problem solving, while classical stress research from Hans Selye onward describes how organisms respond to threats through generalized physiological activation. Information overload research, in turn, has documented performance breakdowns under excessive input, but has largely treated these as information-processing failures rather than stress phenomena. Asriyan and Gevorgyan argue that this separation has obscured a key possibility: that sustained informational demands act on the body through the same allostatic machinery that governs responses to other chronic stressors.

Central to the framework is the concept of allostasis, the active process by which the body maintains stability through change. Building on work by Bruce McEwen and colleagues, the authors distinguish acute adaptive responses from allostatic load, the cumulative wear and tear that results when adaptive responses are repeatedly engaged or poorly regulated. They propose that informational stress operates in this same register: when informational demands are persistent, unpredictable, or perceived as uncontrollable, the regulatory effort they require can accumulate into a physiological burden comparable to that produced by other chronic stressors.

A distinctive feature of the proposal is its insistence that informational stress is not simply proportional to the amount of information received. Drawing on information theory, the authors argue that properties such as uncertainty, entropy, and the rate of change of incoming signals determine how much regulatory work a given stream of information demands. Two environments with the same volume of data can impose very different loads depending on their structure, and the authors suggest that these structural properties can in principle be quantified using tools from information theory rather than relying solely on subjective reports of overload.

The framework also incorporates appraisal, the evaluative process that determines whether a demand is experienced as threatening or manageable. Drawing on the transactional model of stress developed by Richard Lazarus and Susan Folkman, Asriyan and Gevorgyan propose that the same informational stream can be experienced as stimulating or as stressful depending on perceived control, relevance, and available coping resources. This appraisal step explains why informational stress is not a uniform response to volume alone, and why individual and contextual differences matter so much in how overload is experienced.

To connect these ideas to measurable outcomes, the review outlines a set of potential indicators spanning multiple levels. At the cognitive level, these include attentional narrowing, working memory strain, and decision fatigue. At the affective level, they include anxiety, irritability, and the feeling of being overwhelmed. At the physiological level, the authors point to established stress markers such as cortisol, heart rate variability, and inflammatory indicators, alongside more specific measures of cognitive effort such as pupil dilation and neural markers of cognitive control. The framework proposes that these levels are linked through the allostatic process, so that sustained informational demands leave traces across all of them.

The authors are careful to distinguish informational stress from adjacent concepts. It is not the same as cognitive load, which describes momentary demands on working memory during specific tasks. It is not the same as information overload, which describes a condition of excessive input. And it is not the same as digital stress or technostress, which are typically tied to specific technologies. Instead, informational stress is defined by its process character: it emerges when informational demands are sustained, appraised as taxing, and met with regulatory responses that themselves carry costs. This positioning is intended to make the concept useful across contexts, from workplace settings to education to the design of information systems.

The review also addresses recovery, a dimension the authors argue has been neglected in overload research. If informational stress operates through allostatic mechanisms, then recovery periods are not merely pauses but active parts of the stress process, during which physiological systems return toward baseline. The framework suggests that insufficient recovery between informational demands may be a key mechanism by which overload translates into lasting strain, and it calls for research designs that track these dynamics over time rather than capturing overload only in single snapshots.

Asriyan and Gevorgyan frame their contribution as a bridge rather than a replacement. They do not claim that informational stress is a wholly new phenomenon, but rather that existing findings from cognitive load research, stress physiology, and information theory can be integrated into a single process-level account that generates testable predictions. Among these are predictions about when informational demands will produce physiological strain, which structural properties of information streams are most likely to drive that strain, and how individual differences in appraisal and coping shape vulnerability.

The practical implications of the framework are potentially wide-ranging. In workplace contexts, it suggests that interventions should target not only the volume of information but also its structure, pacing, and the sense of control workers have over it. In education, it points to the importance of managing informational demands in ways that support rather than overwhelm allostatic regulation. And in the design of digital environments, it offers a rationale for building systems that respect the recovery needs of their users. The authors emphasize that the framework is intended to be tested, refined, and potentially revised, and that its success will be measured by whether it generates research questions that existing models have not.

Subject of Research: A conceptual review proposing informational stress as a framework connecting information overload to allostatic physiological strain

Article Title: Informational stress in cognitive systems: a review and conceptual synthesis

Article References: Asriyan, E., & Gevorgyan, R. (2026). Informational stress in cognitive systems: a review and conceptual synthesis. Frontiers in Network Physiology, 6, Article 1913737. https://doi.org/10.3389/fnetp.2026.1913737

Image Credits: AI Generated

DOI: 10.3389/fnetp.2026.1913737

Keywords: informational stress, information overload, allostatic load, cognitive load, stress physiology, information theory, Frontiers in Network Physiology, Informational, stress, cognitive, systems, review

Linguistic steganography is the art of hiding secret information inside innocuous-looking text. In a typical covert communication scenario, a sender embeds a hidden message into seemingly normal sentences and transmits the resulting stegotext over a public channel, where the intended recipient recovers the payload. Unlike older modification-based approaches that alter existing documents, generative linguistic steganography embeds information directly during text generation, so that the statistical distribution of the stegotext remains close to that of ordinary language model output. This distribution-preserving property has become a foundational principle of provably secure linguistic steganography, because a message that leaves no statistical fingerprint cannot, in principle, be detected.

The new work targets a specific and increasingly important class of text generators: diffusion language models. Rather than predicting tokens one after another from left to right, as autoregressive models do, diffusion models progressively denoise noisy token states over many inference steps. At each denoising step, the model produces a token-level probability distribution for every position in the sequence. The researchers recognized that this iterative process exposes a whole trajectory of intermediate distributions, each one a potential embedding opportunity with its own entropy characteristics. Previous diffusion-based steganography methods, however, largely ignored this trajectory-level structure.

The team’s motivational study revealed the core problem. Analyzing candidate embedding positions across diffusion trajectories under their default settings, they found that embedding opportunities are highly sparse: the majority of denoising timesteps provide no valid candidates at all, and candidate-rich timesteps occur only occasionally. Existing token-level diffusion steganography methods such as STEAD perform embedding independently within each isolated timestep, which means that when a single step lacks enough compatible candidates, otherwise usable positions are simply wasted. This timestep-local design, the authors argue, is the fundamental bottleneck limiting embedding capacity.

TE-CSA’s answer is an accumulate-then-trigger strategy. Instead of consuming candidates immediately, the framework continuously extracts candidate positions from every denoising timestep and stores them in a global pending pool. When the pool contains at least three compatible positions, the framework constructs an embedding group from the three highest-capacity candidates. Because the same secret bit segment is redundantly embedded across the three positions, the effective capacity of each group is determined by its weakest member, the minimum capacity among the three. This group constraint is central to the paper’s theoretical analysis, in which the authors derive a trajectory-level capacity bound and prove that sorting candidates by capacity and grouping them consecutively attains that bound when the complete retained candidate set is available.

The second key ingredient is entropy-aware filtering. From an information-theoretic standpoint, not all candidate positions are equally valuable. Positions whose token distributions are highly concentrated, meaning the model is already nearly certain which token to place there, carry almost no usable uncertainty and act as bottlenecks: if such a low-capacity position joins an otherwise high-capacity group, it caps the group’s payload. Computing full Shannon entropy for every candidate would be expensive, so the framework uses a min-entropy-inspired approximation based on the maximum token probability, which requires only identifying the most probable token in each local distribution. Candidates whose estimated capacity falls below a threshold are removed before group construction. The strongest filtering configuration, TE-CSA-τ3, retains only positions with capacity of at least 4.

Embedding itself proceeds through distribution-preserving probabilistic sampling. The secret bit sequence is mapped into a continuous value in the interval from 0 to 1 using binary fraction encoding, and the token probability space at each candidate position is partitioned into cumulative intervals. The value falls into one interval, and the corresponding token is sampled under constraints that keep the selection consistent with the original diffusion distribution. Crucially, the authors emphasize that cross-step aggregation and entropy filtering only decide which candidate positions participate in embedding; they do not modify the token probabilities themselves. The security property is therefore inherited from the underlying interval-based embedding mechanism, a point the authors frame as conditional distribution-preserving security rather than unconditional provable security. During extraction, the receiver reconstructs the same candidate sets, filtering decisions, groups, and intervals to recover the message without altering the received text.

The experimental results are striking. Under the default generation setting of length 256, temperature 1.3, and top-p 0.98, the STEAD baseline achieved only 25.30 bits of embedding capacity, while TE-CSA variants reached 469.90 bits without filtering and 470.50 bits with the strongest filtering, an 18.60-fold improvement. The embedding rate rose from 0.09 to 1.84 bits per token, a 20.40-fold gain. Importantly, the improvement does not come from creating new entropy but from converting a far larger fraction of the embedding opportunities already scattered across the diffusion trajectory into effective embedding groups. Runtime overhead remained modest: encoding took 51.53 seconds for TE-CSA-τ3 versus 49.76 seconds for STEAD, because the framework requires no extra denoising iterations or auxiliary neural networks.

Quality and detectability held up as well. Measured by relative perplexity variation, the no-filter variant perturbed generation by 11.20 percent, but entropy-aware filtering cut that figure in half to 5.60 percent while leaving capacity essentially unchanged, validating the claim that low-entropy positions introduce disproportionate distortion for minimal gain. In steganalysis tests against three neural detectors, FCN, R-BiLSTM-C, and BiLSTM-Dense, detection error rates for TE-CSA-τ3 hovered near the 50 percent random-guessing level, at 50.10, 50.06, and 51.38 percent respectively, suggesting the evaluated detectors could barely distinguish stegotext from normal diffusion-generated text. The authors caution that these results characterize practical resistance under the selected detectors, not universal undetectability, and note that LLM-based steganalysis remains a future evaluation frontier.

The study also situates TE-CSA among competing approaches. GTSD, an earlier diffusion-based method, embeds information through prompt mapping and batch selection, with reported rates of roughly 1.56 to 3.94 bits per token depending on configuration, while CascSamp achieves 3.42 bits per token on an autoregressive backbone. TE-CSA reaches a comparable order of magnitude within the more constrained token-level diffusion setting, without requiring pre-shared prompt tables or external mapping resources. The authors acknowledge limitations: the framework depends on trajectories with sufficient candidate diversity, its fixed thresholds may be suboptimal under dynamic generation conditions, and adaptive parameter selection could further improve the capacity-quality-efficiency trade-off. Future directions include extending the approach to other diffusion architectures, longer texts, and multimodal diffusion generation, as well as testing against more advanced steganalyzers.

Subject of Research: High-capacity distribution-preserving linguistic steganography in diffusion language models using trajectory entropy-aware cross-step aggregation

Article Title: TE-CSA: trajectory entropy-aware cross-step aggregation for improving embedding capacity of distribution-preserving linguistic steganography

Article References: Deng, R., Yi, X., Zhang, H., Cao, Y., & Liu, C. (2026). TE-CSA: trajectory entropy-aware cross-step aggregation for improving embedding capacity of distribution-preserving linguistic steganography. Cybersecurity, 9(1), Article 238. https://doi.org/10.1186/s42400-026-00668-2

Image Credits: AI Generated

DOI: 10.1186/s42400-026-00668-2

Keywords: linguistic steganography, diffusion language models, information hiding, entropy, embedding capacity, distribution preservation, cybersecurity, text generation, steganalysis, information theory, covert communication, natural language processing

Alpha-hemolysin is an exotoxin secreted by the bacterium Staphylococcus aureus. In its active form, the protein is built from 293 water-soluble amino acids that assemble into a heptameric pore, a seven-part barrel that punches through cell membranes. That structure, first resolved in crystallographic detail in the 1990s, has made alpha-hemolysin the workhorse of stochastic sensing: the first and still most widely used biological nanopore for detecting individual molecules. When a single pore is embedded in an artificial membrane bathed in electrolyte and a voltage is applied, a steady ionic current flows through the opening. When an analyte molecule wanders into the pore, it partially obstructs that current, producing a characteristic downward blip. Each blip corresponds to one molecule, and the pattern of blips over time carries information about the analyte’s identity and concentration.

In the new experiments, the team used monodisperse polyethylene glycol of molecular weight 1294, abbreviated PEG 1294, as the analyte. The experimental platform was a solvent-free planar lipid bilayer assembled by the classic Montal and Mueller technique: two monomolecular lipid films were apposed across a hole roughly 100 micrometers in diameter in a Teflon partition, forming a barrier that separated two identical compartments of a Teflon chamber. The synthetic lipid 1,2-diphytanoyl-sn-glycero-3-phosphocholine was used at 2 percent by weight in hexane, and the hole was pretreated with a dilute hexadecane solution to promote stable bilayer formation. Both compartments held a concentrated electrolyte of 4 molar potassium chloride buffered with 5 millimolar Tris-citric acid at pH 7.5. Alpha-hemolysin was added to one side, the cis side, at a concentration tuned so that exactly one pore inserted into the membrane.

The researchers then added PEG 1294 to the opposite, trans side at two concentrations, 400 and 1000 micromolar, and applied voltages ranging from +20 to +100 millivolts in 20 millivolt steps. Ionic currents were recorded with an Axopatch 200B amplifier in voltage clamp mode, using silver/silver chloride electrodes connected through agar salt bridges. The signals were low-pass filtered at 15 kilohertz and digitized at a sampling rate of 250 kilohertz, all at a controlled temperature of 25 degrees Celsius. Under these conditions, the current trace showed the textbook signature of stochastic sensing: stepwise transitions between a fully open level and a partially blocked level, each step corresponding to the entry or exit of a single PEG molecule from the pore lumen.

The novelty of the study lay not in the electrophysiology but in the analysis. Rather than focusing on conventional metrics such as event duration or blocking depth, the team turned to Lempel-Ziv complexity, a measure introduced in 1976 by Abraham Lempel and Jacob Ziv to quantify the randomness of finite sequences. The method works by converting a time series into a binary string: each data point is compared with the average of the entire series, receiving a 1 if it exceeds the average and a 0 if it falls below. The resulting string of ones and zeros is then scanned from left to right, and a counter increments every time a new, previously unseen substring appears. Finally, the counter is normalized by a theoretical upper bound, n divided by the base-2 logarithm of n, yielding a complexity value that ranges from 0 to 1 and is independent of sequence length.

The interpretation of this number is intuitive. A value close to 1 indicates a highly random, unpredictable signal with little internal self-similarity, while a value close to 0 indicates a repetitive, ordered series. Lempel-Ziv complexity has previously found applications in biomedical signal analysis, including studies of electroencephalogram background activity in Alzheimer’s disease patients and the detection of ventricular tachycardia and fibrillation in cardiac recordings. Applying it to nanopore data, however, allowed the researchers to ask a question that conventional event statistics cannot easily answer: how complex, in an information-theoretic sense, is the pattern of molecular blocking events?

The answer depended strongly on how much analyte was present. At the lower PEG concentration of 400 micromolar, complexity values were consistently lower than at 1000 micromolar, across every voltage tested. The full range of measured values stretched from a minimum of 0.45, recorded at 400 micromolar and +40 millivolts, to a maximum of 0.97, recorded at 1000 micromolar and +80 millivolts. The physical explanation is straightforward: higher analyte concentrations produce more frequent blocking events, filling the time series with rapid alternations between open and blocked states and pushing the complexity measure toward its upper limit. At lower concentrations, longer stretches of open-pore current dominate, and the signal becomes more ordered and less random.

To probe whether the apparent order at low concentration reflected genuine structure or simply sparse sampling, the researchers performed a clever control. They segmented the original time series recorded at 400 micromolar, shuffled the segments into a random order, and recalculated the complexity of the resulting randomized series. For every applied potential, the shuffled data yielded higher Lempel-Ziv values than the original recordings. This result is diagnostic: if the original series had already been maximally random, shuffling it would have changed nothing. The fact that randomization increased complexity means the unshuffled sequences contained residual structure, and that the biosensor’s output at low analyte concentration falls short of full stochastic behavior.

The team also compared series built exclusively from blocked periods with series built exclusively from unblocked periods, at both concentrations and all voltages. Statistical analysis using Student’s t-test, with a significance threshold of p less than 0.05, found no meaningful differences between the complexity of the blocked-only and unblocked-only series, nor between those segmented series and the original combined recordings. In other words, the complexity of the signal does not reside preferentially in either the moments when PEG occupies the pore or the intervals between occupations; the stochastic character is distributed across the entire record.

The broader implications reach into a long-standing debate in biology about the role of randomness. Molecular biology has traditionally leaned on deterministic principles, yet stochasticity is now recognized as a genuine feature of living systems rather than mere noise. Ion channel kinetics are widely treated as stochastic events, and some theorists argue that intrinsic randomness in the nervous system may even enable flexible decision-making. As the authors note, biological studies often lack the tools to distinguish chaotic, noisy, deterministic, and probabilistic behavior. Lempel-Ziv complexity offers one such tool. Their conclusion is nuanced: the alpha-hemolysin biosensor behaves in a purely stochastic fashion only at higher analyte concentrations, while at lower concentrations its stochasticity diminishes, though it can be mathematically restored by randomizing the event sequence. For engineers designing nanopore sensors, that finding matters, because the reliability of single-molecule detection depends on understanding when the digital current signature is truly random and when it carries hidden structure that smarter algorithms might exploit.

Subject of Research: Stochastic analysis of alpha-hemolysin nanopore ionic current blocking events using Lempel-Ziv complexity

Article Title: Analyzing alpha-hemolysin nanopore behavior using Lempel-Ziv complexity

Article References: Neves, G. F., Machado, D. C., Consoni, L. H. A., Costa, E. V. L., Carneiro, C. M. M., Rodrigues, C. G., & Nogueira, R. A. (2026). Analyzing alpha-hemolysin nanopore behavior using Lempel-Ziv complexity. Discover Chemistry, 3(1), Article 571. https://doi.org/10.1007/s44371-026-01022-8

Image Credits: AI Generated

DOI: 10.1007/s44371-026-01022-8

Keywords: alpha-hemolysin, nanopore, Lempel-Ziv complexity, stochastic sensing, polyethylene glycol, ion channel, planar lipid bilayer, biosensor, information theory, single-molecule detection, biophysics, stochasticity

The central claim of the paper is that conventional approaches to shrinking neural networks get the timing wrong. Post-training quantization, the standard industrial technique, first trains a network in full floating-point precision and then rounds its weights down to lower-precision formats after the fact. Scalar quantization methods treat each weight independently, ignoring the geometry that connects parameters to one another. Lavicka argues that both strategies neglect what he calls the topological limits of optimal information routing during the active learning phase. Instead of discretizing after training, his framework projects the network’s continuous parameter states onto the Leech lattice dynamically, throughout optimization, so that the model lives inside a vector-ternary logic from the moment of initialization.

The choice of lattice is not arbitrary. The Leech lattice, denoted Lambda-24, is a legendary structure in mathematics: a 24-dimensional arrangement of points that achieves the densest possible packing of spheres in that space, with no vectors of short length and an extraordinarily rich symmetry group. Each lattice point, or code atom, sits at a fixed distance from the origin, and the lattice contains exactly 196,560 such minimal vectors. Decoding a continuous vector to the lattice means finding the nearest atom, a task that corresponds to assigning the vector to the Voronoi cell it falls within. These Voronoi cells, the regions of space closest to each lattice point, are highly symmetric, and the paper leans on that symmetry for one of its most striking claims.

According to the study, the symmetric Voronoi cells of the Leech lattice act as optimal geometric filters during training. When stochastic gradient descent computes an update, the signal it carries is a mixture of coherent structural momentum, the meaningful direction in which the network’s parameters should evolve, and isotropic noise, the random jitter that arises from mini-batch sampling and other sources of stochasticity. Because the Voronoi cells partition space symmetrically in all directions, the quantization step rejects noise components that point equally in every direction while preserving the coherent component that carries the network’s learning signal. In this picture, the rigid discrete constraint is not a handicap imposed on training but an active filter that shapes it, grounded in the framework of rate-distortion theory, the branch of information theory that quantifies how much a signal can be compressed before its fidelity degrades beyond a chosen threshold.

The compression figures that emerge from this analysis are dramatic. Lavicka demonstrates mathematically that coupling a network’s weights natively to the Leech lattice compresses the continuous parameters to a theoretical density of 0.76 bits per parameter, well below even a single bit. The argument rests on a two-axiom foundation from which the codebook of lattice atoms is derived, together with an exact Voronoi decoding rule. Because the codebook has a fixed, finite set of atoms and the decoding rule is closed-form, the entire weight structure of a trained network can be stored as a list of integer indices into the lattice, each index requiring only 18 bits to address the 196,560 possible minimal vectors plus the zero vector.

To show that the idea is more than an elegant abstraction, the paper presents a proof-of-concept implementation trained on commodity GPU hardware. Two transformer models, named Pollux-1152 and Pollux-1920, were trained for 10,000 steps within the lattice-constrained space. The smaller model has a 287-million-parameter backbone embedded in a 404-million-parameter artifact, while the larger has a 796-million-parameter backbone within 991 million total parameters. When packed for deployment, the smaller model’s backbone occupies just 27 megabytes on disk and the larger’s just 76 megabytes, both at the predicted 0.76 bits per parameter. The architecture’s dimensions are dictated by the lattice itself: every hidden dimension must be an integer multiple of 24, so attention heads are exactly 24 units wide, embedding widths are 1,152 or 1,920, and even the vocabulary of 50,257 tokens is padded to 50,688 so that it divides cleanly by 24.

Notably, the training runs required no learning-rate schedule, no warmup phase, and no weight decay. The kinetic coefficients that govern how parameters move through the discrete state space are fixed before training by the geometry of the lattice itself, a property the author describes as endogenous kinetic coupling. The empirical results show stable training convergence within this rigidly constrained discrete space, suggesting that the topological bottleneck does not prevent learning, although it does change the character of what the models learn.

That change is visible in the models’ text generation. In qualitative examples included in the paper, both Pollux checkpoints produce text with strong structural fluency: grammatical constructions are consistent, morphological agreements are correct, and rhetorical transitions are coherent. Yet the models are factually unconstrained, inventing linguistically plausible but scientifically nonexistent concepts such as endolymphadoproteins or plasmid plates when asked about mitochondria or plate tectonics. The author attributes this to the topological bottleneck delaying the accumulation of high-entropy factual information, and frames it as consistent with the intended behavior of a stateless cognitive architecture, a design philosophy in which the network captures structure and syntax rather than a memorized store of facts.

There is an important engineering caveat. Modern GPU tensor cores are optimized for dense floating-point arithmetic and cannot natively parse the packed 18-bit combinatorial indices that the lattice encoding produces. The reference implementation therefore materializes decoded lattice vectors into dense half-precision weight matrices at each forward pass, a step that is functionally exact, reproducing bit-identical decoding results, but which does not yet deliver the runtime memory-bandwidth savings that the 0.76-bit code rate promises. Realizing that dividend, the paper notes, would require a native gather-lookup execution substrate, a hardware design that the author has partially covered with a pending international patent application on implementation-level software and hardware aspects, while explicitly leaving the underlying mathematical framework open for independent replication.

The study is careful to position itself as a proof of concept rather than a finished technology. Its stated aim is to explore the fundamental structural principles and theoretical limits of sub-1-bit neural discretization, providing initial empirical evidence that stable syntactic convergence can emerge naturally from rigid topological constraints. If the framework holds up under independent scrutiny, the implications reach beyond compression. Memory-bound hardware, in which the cost of moving data dominates the cost of computing it, is a central bottleneck of modern AI, and a scheme in which weights are natively discrete, geometrically structured, and addressable in under a bit per parameter offers a theoretical foundation for a new class of devices and, perhaps, for discrete cognitive architectures that treat intelligence itself as a lattice-constrained dynamical system. Whether the Leech lattice’s legendary symmetry proves to be the right geometry for machine learning at scale remains an open question, but the paper makes a provocative case that the densest packing in 24 dimensions may be more than a mathematical curiosity.

Subject of Research: Native vector-ternary neural network training via discrete projection onto the 24-dimensional Leech lattice

Article Title: Artificial neural networks as discrete complex systems: native vector-ternary training via the leech lattice (\Lambda _{24})

Article References: Lavicka, A. (2026). Artificial neural networks as discrete complex systems: native vector-ternary training via the leech lattice $$\Lambda _{24}$$. Complex &amp; Intelligent Systems. https://doi.org/10.1007/s40747-026-02517-8

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02517-8

Keywords: Leech lattice, neural networks, quantization, rate-distortion theory, sphere packing, vector quantization, discrete complex systems, information theory, model compression, transformers, cognitive architectures, stochastic gradient descent

The core idea behind the original SABeDM was to track the behavior of a streaming classifier using a beta distribution, a flexible probabilistic description of error rates that lives naturally between zero and one. As data arrives, the model slides a window across the stream and adapts its distributional estimate of how well the learner is performing. When the picture of performance changes sharply, that is a signal that the underlying data distribution may have changed too. The approach already showed promise in earlier work, but like all window-based detectors it faced a familiar dilemma: windows that are too sensitive drown the system in false alarms caused by ordinary noise, while windows that are too forgiving let genuine drifts slip by unnoticed until accuracy has already collapsed.

The new framework attacks that dilemma from two directions at once. The first is information-theoretic. Rather than relying only on summary statistics of the error stream, the enhanced model, denoted SABeDM with a superscript kl, computes the Kullback–Leibler divergence between the probability distributions estimated from consecutive data windows. KL divergence, a foundational quantity in information theory, measures how many extra bits of information are lost, on average, when one distribution is used to approximate another. In practical terms, it gives the detector a single, principled number that quantifies how different the recent past looks from the present. Small divergences correspond to business as usual; large ones flag a genuine redistribution of the data. Because the measure compares full distributions rather than just means or variances, it can pick up both abrupt ruptures and subtle, gradual shifts that simpler statistics tend to smooth over.

The second innovation is temporal. The authors introduce tolerance time, a grace period that allows the detector to absorb minor delays and short-lived fluctuations without immediately declaring a drift. In a live data stream, brief disturbances are ubiquitous: a batch of noisy measurements, a temporary outage, a transient spike in traffic. A detector that reacts to every one of these burns computational resources and, worse, triggers unnecessary model retraining, which can itself degrade performance. Tolerance time introduces deliberate patience into the system, requiring that evidence of a distributional shift persist for a defined interval before an alarm is raised. The combination is elegant in its symmetry: KL divergence supplies sensitivity to real change, while tolerance time supplies immunity to phantom change.

To find out whether this dual mechanism actually delivers, the team evaluated the framework on six benchmark datasets spanning the standard menagerie of drift scenarios: SEA_a, SEA_g, MIXD, HYP, PHI and WET. These benchmarks are deliberately engineered to stress different failure modes, from sudden abrupt concept changes to gradual and incremental drifts, and the WET dataset adds a real-world, high-dimensional test case where clean theoretical behavior often falls apart. The proposed detector was pitted against a roster of state-of-the-art baselines that includes Streaming Random Patches (SRP), ADWIN, DDM and EDDM, methods that have anchored the drift detection literature for years and, in the case of ADWIN and DDM, remain default choices in many production pipelines.

The results were consistent and, in several cases, striking. On the SEA_a dataset, the enhanced SABeDM achieved an accuracy of 83.50 percent and an F1-score of 82.75 percent, compared with 73.68 percent and 73.87 percent respectively for SRP, the strongest competitor in that comparison. On the more complex PHI dataset, the framework reached 96.10 percent accuracy and a 96.15 percent F1-score, outperforming every baseline tested. Gains extended across precision and recall as well, indicating that the improvements were not an artifact of trading one error type for another but reflected a genuinely better-calibrated detector. The authors also report minimal detection latency, meaning the system tends to notice drift soon after it begins, which is critical because every example processed between the onset of drift and its detection is an example a deployed model may classify wrongly.

Perhaps the most persuasive evidence came from the real-world WET dataset, where the gap between theory-friendly benchmarks and messy practice usually narrows dramatically. There, the enhanced framework lifted the F1-score to 69.00 percent from 60.19 percent for SRP, a gain of nearly nine percentage points in a setting where such margins are rare. The authors attribute this robustness to the interplay of the two new components: KL divergence detects the fine-grained distributional texture of real data, while tolerance time prevents the noise inherent in real streams from overwhelming the detector. The framework also held up in high-dimensional settings, a known weak point for many statistical detectors whose assumptions strain as dimensionality grows.

The significance of the work extends beyond one benchmark table. Concept drift detection sits at the foundation of trustworthy machine learning in dynamic environments, from fraud monitoring and network intrusion detection to predictive maintenance and adaptive recommendation. When drift goes undetected, models fail silently, and the cost is measured in bad decisions rather than error bars. When detection is too twitchy, systems churn through retraining cycles and lose the stability that operators depend on. By grounding the detection decision in a rigorous information-theoretic quantity and pairing it with an explicit model of temporal tolerance, the new framework offers a principled middle path, one that other researchers can extend to related problems such as detecting recurrent concepts, distinguishing real from virtual drift, and handling drift in image and other complex data streams, areas the authors and a growing survey literature identify as open challenges.

There are, of course, caveats worth keeping in mind. The reported evaluations rest on six datasets, and the authors note that no new datasets were generated or analyzed beyond those used in the study, so independent replication on additional industrial streams will be the natural next test. The choice of tolerance time parameters will also matter in practice, since the right grace period likely depends on the application’s tolerance for delayed detection versus false alarms. Still, the direction is clear and the evidence is strong. As data streams grow faster and models are asked to learn continuously in environments that refuse to stand still, detectors that can both sense the faintest whisper of change and ignore the routine noise of the stream will only become more valuable. This study suggests that an old idea from information theory, applied inside an adaptive probabilistic window with a well-timed pause, may be exactly the combination the field has been waiting for.

Subject of Research: Concept drift detection in data streams using KL divergence and tolerance time within sliding adaptive beta distribution models

Article Title: Information-theoretic drift detection using KL divergence and tolerance time in Sliding Adaptive Beta Distribution Models

Article References: Bian, W., Ature, A., Yong, C. H., & Rabiu, S. (2026). Information-theoretic drift detection using KL divergence and tolerance time in Sliding Adaptive Beta Distribution Models. Knowledge and Information Systems, 68(1), Article 273. https://doi.org/10.1007/s10115-026-02893-0

Image Credits: AI Generated

DOI: 10.1007/s10115-026-02893-0

Keywords: concept drift, KL divergence, SABeDM, data streams, machine learning, beta distribution, tolerance time, drift detection, streaming data, information theory, ADWIN, SRP

The team turned to one of the most intensively studied systems in developmental biology: the fruit fly embryo. In the early Drosophila embryo, maternal morphogens lay down the initial coordinates along the anterior-posterior axis, the head-to-tail line of the body. These inputs drive a network of four gap genes, knirps, Krüppel, hunchback, and giant, whose broad expression domains in turn pattern the striped expression of pair-rule genes such as even-skipped, paired, and runt. Those stripes ultimately map out the segmented body plan of the adult fly. Because gene expression in these embryos is extraordinarily reproducible between individuals, with the variance of stripe positions measuring roughly half the distance between adjacent nuclei, the system offers an ideal testing ground for quantitative, information-theoretic analyses of how cells decode their position.

Earlier work had established that gap gene expression carries enough information for a cell to estimate its position along the anterior-posterior axis with an accuracy of about one percent of the embryo’s length. That sounds extraordinarily precise, and it is, but a recent theoretical analysis revealed a paradox. One percent precision is not sufficient to uniquely determine a cell’s position among the roughly ninety nuclei that line the axis. In principle, two adjacent cells could decode their positions in the wrong order, swapping identities. Researchers proposed that long-range correlations in gene expression along the entire axis could resolve this so-called information gap, because such correlations constrain the space of possible collective expression patterns and thereby reduce how much information each individual cell needs. But that raised an obvious biological objection: no single cell can plausibly access information about expression patterns far away on the other side of the embryo.

The new study resolves this tension with an elegant insight. The authors showed mathematically that the long-range correlations invoked in previous work can emerge entirely from correlations between immediately adjacent cells. When neighboring positions along the axis have highly correlated expression levels, these local relationships propagate to generate the embryo-wide correlation structure, and above a critical threshold they close the information gap. Crucially, the required threshold corresponds to pairwise correlations between adjacent nuclei exceeding 0.95, a distance across which cells can plausibly interact, whether through direct contact, diffusing molecules, or shared mechanical cues. When the researchers examined experimental measurements of gap gene and pair-rule gene expression in wild-type fly embryos, they found that pairwise correlations between adjacent positions did indeed exceed 0.95 across the full length of the axis.

Building on this observation, the team formulated a neighborhood-informed decoder, a probabilistic framework that estimates a cell’s position using not only the cell’s own gene expression but also the expression levels of its immediate neighbors. Formally, the framework compares a cell-independent decoder, which has access only to the expression profile g of a single position, with a neighborhood-informed decoder, which also receives the expression profiles of flanking positions. Under standard assumptions that gene expression fluctuations follow a Gaussian distribution, the authors derived a closed-form expression for the positional error in each case and proved a key result: adding neighborhood information can never increase the positional error, and in practice it strictly decreases it. The amount of positional information, measured in bits, is inversely related to that error, so shrinking the error means gaining information.

The quantitative payoff was dramatic. Applying the neighborhood-informed decoder to fluorescence measurements from wild-type embryos, the researchers found that the standard deviation of predicted position dropped to roughly 0.2 percent of embryo length, nearly an order of magnitude smaller than the roughly one percent achieved by cell-independent decoding. In information terms, the neighborhood-informed decoder extracted an average of 7.0 bits of positional information per cell, compared with 4.4 bits for the cell-independent decoder. Since uniquely specifying a cell’s location along the axis requires about 6.49 bits, the neighborhood-informed decoder clears the bar while the cell-independent decoder falls short. In other words, the microenvironment contains exactly the information that individual cells appear to lack.

The framework also proved its worth as a predictive tool. Trained on hundreds of embryos, the neighborhood-informed decoder produced posterior distributions over positions that were more sharply peaked around the true locations, with lower prediction errors and lower ambiguity than the cell-independent alternative, and the improvements were statistically significant across virtually all embryo-to-embryo comparisons. The decoded positions could then be used to reconstruct downstream pair-rule gene expression patterns, and here too the neighborhood-informed approach won decisively, yielding significantly more accurate reconstructions of the even-skipped, paired, and runt stripe patterns across the axis. The advantage persisted even in mutant embryos whose maternal signals, such as Bicoid and Nanos, had been genetically perturbed, producing altered gap gene landscapes and shifted stripes. Even there, decoding with neighborhood information reduced the uncertainty of inferred positions and better predicted the mutant pair-rule patterns, suggesting that the mapping from gap gene expression to positional readout genuinely exploits local context rather than isolated cellular measurements.

To test whether the principle extends beyond the fly, the authors applied the same analysis to two mammalian systems. The first was the gastruloid, an in vitro model of early mammalian development in which mouse embryonic stem cells self-organize into elongated structures with anterior-posterior patterning. Analyzing the germ-layer markers Bra, Cdx2, FoxC1, and Sox2 along the gastruloid axis, the team found that cell-independent decoding again fell short of the roughly 5.2 bits needed for unique localization, delivering only about 4.6 bits on average, while the neighborhood-informed decoder reached 6.8 bits, comfortably above the requirement. The second system was the developing mouse neural tube, where the antiparallel signaling gradients of Sonic hedgehog and bone morphogenetic proteins pattern neural progenitors along the dorsal-ventral axis. Once more, neighborhood-informed decoding supplied enough information, about 6.1 bits against a requirement of 5.4 bits, to uniquely specify positions along 89 percent of the axis, whereas cell-independent decoding was insufficient along most of it. The consistency across fly embryos, synthetic mammalian tissues, and the neural tube suggests that neighborhood-based positional decoding may be a general feature of developmental patterning rather than a quirk of one model organism.

The authors are careful about what their framework does and does not claim. The analysis is deliberately agnostic about mechanism: information could flow between neighbors through direct cell-cell contacts, diffusing signals, altered membrane mechanics, or adhesive molecules, and the information-theoretic results hold regardless of which channel is used. The study also cannot determine whether the spatial correlations were already baked into the cells when the data were collected, or whether neighborhood interactions actively help cells refine their positions as development proceeds; perturbation experiments targeting cell-cell interactions would be needed to distinguish these scenarios. The framework additionally requires data aligned along a single well-defined axis with enough replicates to estimate covariance matrices, which is why the authors focused on these three well-characterized systems. Future extensions could move to two- and three-dimensional contexts and to spatial transcriptomics data, potentially using graph-based definitions of neighborhoods that go beyond simple physical adjacency.

Even with those caveats, the implications are broad. The work reframes a foundational question in developmental biology: instead of asking only what a cell knows on its own, it asks what a cell and its companions collectively know. The finding that local microenvironments carry the missing bits needed for unique identity specification offers a satisfying resolution to the positional information gap and provides a quantitative tool that can be applied wherever reproducible patterning emerges. It also hints at why tissues are so robust: a decoding strategy that pools information across neighbors is inherently resilient to the noise and variability that plague individual measurements. As researchers continue to map the conversations cells hold with their surroundings, this study suggests that the answers to how an embryo builds itself may be written not in any single cell, but in the shared language of the neighborhood.

Subject of Research: Neighborhood-informed positional information and cell identity specification during embryonic development

Article Title: Neighborhood-informed positional information for precise cell identity specification

Article References: Erez, M., Friedman, R., & Nitzan, M. (2026). Neighborhood-informed positional information for precise cell identity specification. Molecular Systems Biology, 22(7), 1118-1131. https://doi.org/10.1038/s44320-026-00211-y

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00211-y

Keywords: positional information, information theory, Drosophila embryo, gap genes, pair-rule genes, morphogen gradients, cell fate, gastruloids, neural tube, developmental biology, gene expression decoding, microenvironment

The study, published in the International Journal of Data Science and Analytics by Grace Kim of Arizona State University and Dongyung Kim of Benedictine University, introduces a modified metric called Atwood-weighted Normalized Mutual Information, or ANMI. The name points to its inspiration: the Atwood number, a dimensionless quantity from fluid dynamics that describes the instability of a interface between two fluids of different densities when the heavier one is accelerated into the lighter one. First formalized by Geoffrey Taylor in his landmark 1950 analysis of liquid surface instability, the Atwood number determines whether small perturbations at a fluid boundary grow into the dramatic, mushroom-shaped fingers known as Rayleigh-Taylor instabilities. The authors argue that an analogous instability afflicts standard NMI when the information density of a model’s predictions diverges sharply from that of the ground truth.

The core problem the researchers target is over-clustering, a scenario in which a model partitions data into far more groups than actually exist, or otherwise produces clusters whose statistical density differs dramatically from the true class structure. In such cases, standard NMI can yield over-optimistic results, reporting high agreement between predictions and ground truth even when the model has failed to match the intrinsic information content of the dataset. This is not merely a theoretical quibble. Recent work cited in the paper, including a 2025 analysis in Nature Communications by Jerdee, Kirkley, and Newman, demonstrated that normalized mutual information is a biased measure for classification and community detection, lending independent weight to the concern that the field’s favorite yardstick systematically distorts comparisons.

To build their corrective, the authors define what they call an Information Atwood Number, denoted A_I, which is computed from the entropy difference between the ground truth distribution and the predicted distribution of labels. Entropy, in the information-theoretic sense, measures how spread out or uncertain a probability distribution is. A model that lumps nearly all images into a handful of giant clusters has a very different entropy profile from one that spreads predictions evenly across many fine-grained categories, and both may differ from the entropy of the true labels. By taking the difference between these entropy profiles and normalizing it in the manner of the classical Atwood number, the researchers obtain a single value that captures how mismatched the information densities of prediction and reality truly are.

That mismatch value then becomes a penalty term. In the ANMI formulation, the standard normalized mutual information score is weighted by a complexity-aware factor derived from the Information Atwood Number. When a model’s predicted distribution closely matches the entropy profile of the ground truth, the penalty is mild and ANMI behaves much like ordinary NMI. But when the densities diverge sharply, as in over-clustering scenarios where a model fragments the data into many sparsely populated groups, the penalty grows and drags the score down. The result, according to the authors’ experimental results, is a metric that provides a more robust and conservative assessment than standard NMI, particularly penalizing models that fail to match the intrinsic information density of the data they are meant to classify.

The physics analogy is more than a naming flourish, the authors suggest. In Rayleigh-Taylor instability, the Atwood number determines the growth rate of perturbations: when the density contrast is small, the interface remains nearly stable, but when it is large, even tiny ripples amplify explosively. Similarly, the researchers argue, small discrepancies between the entropy of predictions and the entropy of truth are benign and barely affect evaluation, whereas large discrepancies signal a fundamental instability in the model’s representation of the data, one that standard NMI fails to register. By encoding the density contrast directly into the metric, ANMI makes that instability visible in the final score. The approach reflects a broader trend of importing concepts from physical science into machine learning evaluation, where dimensionless ratios and conservation-style arguments can expose pathologies that raw performance numbers conceal.

The development and verification of the metric drew on established information-theoretic foundations. The paper builds on Marina Meilă’s influential 2007 work comparing clusterings with information-based distances, and on Vinh, Epps, and Bailey’s 2010 study of normalization properties and chance correction in clustering comparison measures, both of which established the mathematical ground rules for metrics like NMI. The experimental evaluation was conducted on widely used, publicly available benchmark image datasets, including MNIST, CIFAR-10, and STL-10, the standard proving grounds for classification and clustering algorithms. The authors also reference the Scikit-learn machine learning library, the de facto standard implementation environment for such metrics in the Python ecosystem, suggesting a straightforward path for practitioners who wish to adopt the new measure.

The practical stakes are considerable. Clustering and classification evaluation scores do not merely describe models; they decide which models get funded, deployed, and built upon. If NMI systematically rewards over-clustered representations, then research lines that fragment data excessively may appear more promising than they are, while simpler, better-calibrated models are unfairly disadvantaged. In applications such as biomedical image segmentation, where one of the authors has previously published work on numerical methods for partial differential equations in image segmentation, an inflated agreement score could mask a model’s failure to respect the true informational structure of tissue classes. A conservative metric that refuses to flatter entropy-mismatched predictions could change which algorithms rise to the top of leaderboards.

The paper arrives amid a broader reassessment of how the machine learning community grades itself. The 2025 Nature Communications finding that NMI is a biased measure, together with a 2026 Journal of the American Statistical Association paper on validating internal clustering validation measures, indicates that the problem of evaluation bias is attracting attention from multiple directions. ANMI’s contribution is a concrete, computationally simple weighting scheme that any researcher already computing NMI could extend with an entropy calculation on the label distributions. Because the metric requires no additional data beyond what NMI already uses, its adoption cost is low, though its impact will depend on whether independent groups replicate the authors’ findings that the penalty term improves robustness across diverse datasets and clustering regimes.

For now, the study stands as a reminder that the tools scientists use to measure success are themselves scientific instruments, subject to calibration and correction. The image of two fluids of mismatched density, one collapsing catastrophically into the other, turns out to be an apt metaphor for what happens when a model’s information density and the truth’s information density drift apart: the evaluation interface becomes unstable, and scores that look serene on the surface conceal violent disagreement underneath. Whether ANMI becomes a standard fixture in the clustering toolkit or one correction among several in an evolving debate, its central lesson is already clear. The next generation of machine learning benchmarks may owe as much to the physics of fluids as to the mathematics of information.

Subject of Research: A physics-inspired evaluation metric, Atwood-weighted Normalized Mutual Information, for image classification and clustering assessment

Article Title: Atwood-weighted Normalized Mutual Information (ANMI): a physics-inspired metric for image classification evaluation

Article References: Kim, G., & Kim, D. (2026). Atwood-weighted Normalized Mutual Information (ANMI): a physics-inspired metric for image classification evaluation. International Journal of Data Science and Analytics, 22(1), Article 320. https://doi.org/10.1007/s41060-026-01288-2

Image Credits: AI Generated

DOI: 10.1007/s41060-026-01288-2

Keywords: Normalized Mutual Information, ANMI, Atwood number, image classification, clustering evaluation, information theory, entropy, machine learning, over-clustering, Rayleigh-Taylor instability, fluid dynamics, benchmark datasets

Pearson correlation has long been the workhorse of time-series comparison in neuroscience. As a model-based measure, it operates under the assumption that the two input signals are jointly Gaussian and largely free of outliers. When those assumptions hold, the sample Pearson correlation coefficient is easy to compute from experimental recordings and offers a transparent interpretation: a value near one signals a strong linear relationship, a value near zero signals the absence of a consistent linear one. But the brain rarely cooperates so neatly. Neural evoked responses can be noisy, non-Gaussian, and, crucially, linked by dependencies that go beyond simple linearity. A zero correlation, the authors note, does not even imply the absence of structure, since alternating stretches of positive and negative correlation can cancel out. This is precisely where mutual information, a model-free measure rooted in information theory dating back to Claude Shannon’s foundational 1948 work, offers an attractive alternative capable of capturing both linear and higher-order relationships between signals.

The catch, and the reason mutual information has not displaced correlation in everyday neuroimaging practice, is estimation. For continuous data, there is no simple, universally accepted estimator of mutual information, and accurate estimation from the limited samples typical of neurophysiological recordings is notoriously challenging. Decades of methodological development have produced a crowded toolbox: adaptive binning schemes, kernel density estimators, the Kraskov–Stögbauer–Grassberger nearest-neighbor estimator, Gaussian copula approaches, and even neural network–based estimators. Each carries its own advantages, biases, and tuning parameters. The underlying probability distribution of the data, the choice of estimator-specific parameters, and normalization factors can all dramatically shift the resulting estimates. The authors of the new study, Anni Hukari, Silvia Federica Cotroneo, and Riitta Salmelin of Aalto University’s Department of Neuroscience and Biomedical Engineering, set out to bring order to this landscape for the specific case of neural evoked responses.

The study’s methodological core is a carefully controlled simulation framework. The researchers generated base signals as Morlet wavelets, resembling the oscillatory waveforms common in evoked brain activity, and then subjected pairs of copies to a battery of realistic transformations. They varied the signal-to-noise ratio, injected sparse high-amplitude outliers resembling muscle artifacts, appended noise-only segments to mimic signal cropping, imposed small time shifts, altered response duration, and changed relative magnitude. Each configuration was repeated one thousand times across different noise iterations, allowing the team to map how each estimator’s output behaved across the full parametric space. In parallel, they anchored their interpretation with a novel statistical device: adaptive lower bounds, constructed by comparing each reference signal against one thousand Gaussian noise realizations matched in mean and standard deviation, yielding empirical ninety-nine percent confidence thresholds against which the true comparisons could be judged.

The parameter tuning phase alone yielded practical insights that researchers can apply immediately. For the kernel density estimator, the team adopted Scott’s rule for bandwidth selection, which produced stable performance in both simple and nonlinear comparison scenarios. For the Kraskov estimator, the authors found that while the original developers recommend choosing between two and four nearest neighbors, setting k equal to five widened the gap between genuine signal similarity and noise comparisons, improving separability. For adaptive binning, Doane’s rule correctly identified the optimal number of bins, thirteen in their configuration, corresponding to the peak of the similarity estimate curve. Crucially, the lower bounds proved instrumental in these decisions: small kernel bandwidths, for instance, inflated the noise floor so much that true similarity became indistinguishable from chance, a failure mode that would be invisible without such a reference.

The simulation results revealed a consistent behavioral split. Pearson correlation and the kernel density estimator formed one pair, while the Kraskov and adaptive binning estimators formed another. When signals were strictly shape-identical, Pearson correlation performed reliably, even tolerating remarkably small sample sizes, and it separated signal from noise comparisons slightly earlier, at zero decibels of signal-to-noise ratio, where the mutual information estimators still struggled. Yet when signals shared information without sharing shape, Pearson correlation and the kernel estimator faltered. Under small time shifts, Pearson correlation and the kernel estimator dropped below their noise thresholds almost immediately, while the Kraskov and binning estimators remained above their bounds, detecting the underlying relationship. Similarly, with sparse high-amplitude outliers, Pearson correlation and the kernel estimator degraded rapidly, whereas the nearest-neighbor and binning approaches declined more gracefully. The two estimation families, in short, are sensitive to different properties of the data.

The study also delivered concrete numerical guidance. All estimators required positive signal-to-noise ratios to function meaningfully, but their performance stabilized around twenty decibels, a level at which any added noise causes negligible distortion. Sample sizes of roughly one hundred began to separate genuine similarity from noise, and by two hundred and ten samples, corresponding to a six-hundred-hertz sampling rate for their signals, all estimators converged to stable estimates. Sampling frequencies above two hundred hertz proved sufficient, improvements plateaued near five hundred hertz, and oversampling beyond that added computational cost and redundancy-driven bias without benefit. Perhaps counterintuitively, the researchers recommend deliberately adding a small amount of Gaussian noise to signals, even after preprocessing, when using the Kraskov or kernel density estimators, since near-identical samples can cause these methods to break down entirely.

To demonstrate that these findings survive contact with real data, the team applied all four estimators to magnetoencephalography recordings from a picture-naming task, comparing a reference channel near a known brain activation source with every other sensor in the whole-head helmet array. The behavioral pairing observed in simulations reappeared: Pearson correlation and the kernel estimator produced variable estimates with tight noise bounds, while the Kraskov and binning estimators yielded broader bounds and more uniform similarity across sensors. When the researchers ranked all channels by similarity rather than comparing absolute values, sensors closest to the reference ranked highest for every estimator, consistent with the spatial spread of magnetic sources. But the mutual information estimators additionally assigned high similarity to sensors capturing delayed or slightly distorted versions of the response, which Pearson correlation systematically under-ranked. The ranking-based approach proved essential, because all channels in a single recording share so much information, from common stimulus drive to shared preprocessing artifacts, that comparison against random noise alone provided limited insight.

The study is candid about the interpretive traps that remain. Mutual information is unbounded, and normalization strategies, including the widely used one adopted here, introduce their own upward biases at low values. The choice of reference signal matters enormously: because the reference channel in the case study contained a mixture of neural activity, noise, and artifacts, the mutual information estimators recognized similarities in other channels based on all of these factors, making it difficult to judge whether high similarity reflected genuine neural activation or merely shared residual artifacts such as eye blinks. The authors’ prescription is methodological humility: never interpret similarity values in absolute terms, always establish adaptive lower bounds tailored to the signal’s sample size and variance, and reason in terms of rankings. Their conclusion offers a practical decision rule for the field. When the goal is recognizing the exact same waveform across signals, Pearson correlation remains the simpler, efficient choice. When signals are expected to share features without being identical, the Kraskov estimator, which proved the easiest of the mutual information methods to tune, emerges as the recommended tool, potentially reshaping how connectivity, artifact removal, and stimulus-response analyses are conducted across EEG and MEG laboratories worldwide.

Subject of Research: Estimating mutual information and Pearson correlation for comparing neural evoked responses in EEG and MEG signals

Article Title: Estimating Mutual Information and Pearson Correlation on Neural Evoked Responses

Article References: Hukari, A., Cotroneo, S. F., & Salmelin, R. (2026). Estimating Mutual Information and Pearson Correlation on Neural Evoked Responses. Neuroinformatics, 24(4), Article 61. https://doi.org/10.1007/s12021-026-09784-3

Image Credits: AI Generated

DOI: 10.1007/s12021-026-09784-3

Keywords: mutual information, Pearson correlation, neural evoked responses, magnetoencephalography, electroencephalography, similarity estimators, signal processing, information theory, neuroimaging, time-series analysis, nonlinear dependence, statistics

Surprisal quantifies how unexpected a word is given the context that precedes it, as judged by a probabilistic model. In psycholinguistics, the concept has a long pedigree: work by John Hale in 2001 and Roger Levy in 2008 formalized the idea that the surprisal of a word predicts how hard it is to process, and subsequent studies linked word-level surprisal to reading times and to neural signatures of comprehension such as the N400 brain response. Words with high surprisal are less predictable and typically demand more cognitive effort; words with low surprisal glide by. Modern large language models, trained to predict the next word in a sequence, provide a convenient and powerful engine for estimating these probabilities at scale.

What makes the new work distinctive is its shift in perspective. Instead of using surprisal to probe how readers process a fixed text, the Spanish team asked whether surprisal could characterize the producer of a text. When someone generates language, the average surprisal of their words reflects how predictable—or unconventional—their discourse tends to be. That raises an obvious psychometric question: is mean surprisal a stable property of the person, or merely a quirk of one particular answer that will not generalize to anything else that person writes? If the former, surprisal joins the ranks of individual differences like vocabulary size or processing speed; if the latter, it is little more than a text-level statistic.

To find out, the researchers conducted a series of four studies using datasets from previously published work, reanalyzing language-based responses produced by the same individuals across multiple different items and tasks. Crucially, they treated the problem with the standard toolkit of psychological measurement: reliability analysis, inter-item correlations, and corrections for measurement error. If surprisal were merely a property of individual texts, scores from different writing prompts completed by the same person should be essentially unrelated. If it were a trait, those scores should correlate, just as different items on a well-constructed questionnaire do.

The results were striking. When surprisal scores were averaged across the different language-based items within a task, reliability was excellent—the measure behaved like a coherent psychological scale. Within individual items, reliability varied, but across items the scores converged. In other words, a person who produced highly predictable language on one prompt tended to produce highly predictable language on another, even when the content differed completely. The team also found, however, that surprisal was significantly influenced by the length of the text answer, a methodological caveat that future studies will need to control for when comparing texts of different sizes.

The researchers then examined how mean surprisal scores related to other variables, drawing on datasets that included measures of language skills, domain-general cognitive abilities, and personality traits. The pattern of correlations suggested that surprisal is not an isolated curiosity. It related to certain cognitive functions, consistent with the idea that producing predictable versus unpredictable language taps into underlying processing efficiency. The authors also propose that surprisal may capture alternative styles of speech or communication—broad tendencies in how people assemble their discourse that go beyond what any single answer reveals.

Technically, the measure is closely related to perplexity, a familiar quantity in computational linguistics; the authors note that surprisal scores can be seen as a linear transformation of perplexity but argue that surprisal is more interpretable from a psychological standpoint. The models used were left-to-right next-word predictors, meaning that each word’s surprisal was computed from the words preceding it, mirroring the incremental nature of human language production. Averaging word-level surprisal across a response yields a single predictability score per text, and averaging across texts yields the individual-difference measure at the heart of the study.

The implications reach well beyond psycholinguistics. Automated scoring of essays, the assessment of constructed responses, and the growing field of speech-based biomarkers for neurological conditions all depend on extracting meaningful, stable signals from language samples. If average surprisal tracks a durable characteristic of the speaker or writer, it could complement existing lexical and semantic features in applications ranging from educational assessment to the detection of cognitive decline, where changes in the structure and predictability of spontaneous speech have been linked to conditions such as Alzheimer’s disease and frontotemporal dementia. The new findings provide the psychometric foundation such applications require: evidence that the signal is reliable, not noise.

The study also carries a cautionary message for the rapidly expanding literature that uses large language models to score human language. Any single text-based score is contaminated by measurement error and by item-specific effects, and correlations between such scores and other variables will be attenuated accordingly. The authors show, for example, that correcting for attenuation can substantially raise observed inter-item correlations, meaning that raw correlations should be read as lower bounds. Transparency was a priority for the team: all surprisal measures, code, and supplementary materials are publicly available through the Open Science Framework, allowing other researchers to scrutinize and extend the approach. As artificial intelligence increasingly mediates how human language is measured, this work demonstrates that concepts born in information theory can be tamed into rigorous psychological instruments—provided researchers ask the same questions of them that psychometricians have asked of tests for a century: Is it reliable? Does it generalize? And what, exactly, does it measure about the person behind the words?

Subject of Research: Using large language model surprisal as a stable individual-difference measure of language production predictability

Article Title: Surprisal from Large Language Models as an individual difference: Capturing the individual regardless of the text answer

Article References: Martínez-Huertas, J. Á., Martínez-Mingo, A., López Rosell, A., Kreitchmann, R. S., & Jorge-Botana, G. (2026). Surprisal from Large Language Models as an individual difference: Capturing the individual regardless of the text answer. Behavior Research Methods, 58(10), Article 288. https://doi.org/10.3758/s13428-026-03117-7

Image Credits: AI Generated

DOI: 10.3758/s13428-026-03117-7

Keywords: surprisal, large language models, individual differences, psycholinguistics, language production, reliability, predictability, Behavior Research Methods, psychometrics, information theory, cognitive skills, automated text scoring