Neural Network Interpretability

Neural networks can learn algorithmic input-output mappings, but trusting a learned executor requires more than a correct final answer because the state transitions that produce it are usually hidden. To make those transitions visible, we introduce a trace-supervised symbolic neural CPU, a factorized learned execution architecture that combines recurrent control, an explicit operation router over a fixed differentiable arithmetic-logic unit bank, destination-masked register writeback, complete trajectory supervision and matched fixed-point replay. The model exposes the selected operation, source and destination registers, register trajectory, memory signals and writeback semantics at every step. On the principal 16-wide benchmark, the non-quantized executor reproduces reference execution exactly, while the eight-bit quantization-simulated executor preserves the symbolic operation path through programs of 1,000 instructions. When the same execution is evaluated against a matched fixed-point replay, the residual numerical drift disappears, showing that it comes from a mismatch between continuous and low-precision reference semantics rather than from execution failure. We compare recurrent, Transformer, temporal-convolution, temporal graph-inspired and state-space controllers, and the ablations show that operation-gate supervision is necessary for an inspectable execution path. Hidden-opcode memory-pressure tasks expose the remaining limits in delayed state use and temporal binding. We also extend the interface with ValueMemory, hybrid adaptive leaky integrate-and-fire controllers, candidate-constrained symbolic control trained through behaviour cloning and actor-critic reinforcement learning, and an RV32I base-integer semantic bridge. Together, these results establish a trace-verifiable framework for interpretable, low-precision and controllable neural execution.

The current state of the art in AI/ML rests on deep neural architectures, which, in general, suffer from a lack of interpretability. Interpretability is crucial to gleaning insights while analyzing experimental data, where quantitative predictions may not be adequate for a scientist. We present a three layer neural architecture, SAMPAT (Smooth Approximation via Multivariate Polynomials and Analytic Transformations), that can provably learn a continuous, everywhere differentiable function, that can approximate any smooth function arbitrarily closely. SAMPAT's approximant can be expressed as a closed and compact algebraic, analytic expression, providing complete interpretability. Experiments on synthetic and benchmark datasets indicate that SAMPAT yields competitive performance with simpler representations. For many tasks, a two layer SAMPAT suffices. By imposing restrictions on the connectivity between neurons, SAMPAT may be used to provide a range of approximants, including regular and trigonometric polynomials, rational expressions, Gaussians, mixtures of Gaussians, as well as arbitrary combinations of the same; without restrictions, it learns a suitable structure. SAMPAT may be used to factorize polynomials and model nonlinear systems. With the addition of skip connections, a 4 to 6 layer SAMPAT is adequate to represent a substantive range of methods widely used in AI/ML, allowing the choice of a model's family, not just its parameters, to also be optimized as part of the learning process.

In artificial and biological neural networks, concepts are often encoded as consistent linear directions in representation space. In deep learning, this idea is known as the linear representation hypothesis and underpins many interpretability and control methods based on linear probes, from concept detection to activation steering. Yet while prior work has studied whether such directions should exist $\textit{after}$ training, the dynamics of how they emerge $\textit{during}$ training remain poorly understood. Here, we develop a framework to study the alignment of concept directions during training - a process we call "abstraction". In a minimal linear network setting, we obtain exact solutions for the full trajectory of abstraction. These solutions reveal key analytic principles governing abstraction: (i) data and target geometry jointly determine abstraction at the end-of-learning, (ii) abstraction improves with network depth, and (iii) initialization scale controls the maximum abstraction reached during training. Extending our theory to nonlinear networks, we analyze how the choice of nonlinearity affects abstraction dynamics: erf networks approximate the linear theory, while abstraction in ReLU networks depends less on target geometry and more on input geometry. Across both, we prove a striking attenuation law: both nonlinearities weaken abstraction in activations relative to preactivations. We find evidence for this law in open models (DINOv3, Gemma 4) and apply our theory to improve linear probe generalization in LLMs. Together, our results provide a dynamical theory of abstraction with implications for interpretability and control.

Over the last few years, there has been an increased interest in making machine learning models more interpretable. Although a great deal of effort goes into developing techniques for interpreting the interactions learned by a given model, fewer studies focus on assessing the quality of such explanations. Even fewer focus on how to adjust the model to produce explanations faithful to prior knowledge, a process known as explanation-guided learning. Furthermore, most approaches in this area focus on classification problems and usually assume prior knowledge about which input features or regions are most important. In this work, we introduce a new approach to steering neural networks based on partial dependence, such that their average response to certain features aligns with specific functional domain knowledge about the problem. We empirically demonstrate on a range of regression problems, including dynamical systems forecasting, that models whose training has been controlled using our method perform better than unconstrained models and are more data-efficient. Moreover, we highlight that interpretations obtained from the former actually align with the user-provided knowledge, whereas those obtained from the latter do not.

End-to-end neural audio models achieve high-fidelity compression and generation. We might read that performance as evidence they directly represent interpretable features such as pitch and timbre, but a model can produce plausible outputs without doing so. A model may encode these features in any reachable basis, but regardless of which, the features are well described as compositions of time-frequency-localized primitives. Whether state-of-the-art encoders preserve access to these primitives, and thus to compositions of them, remains unclear. Through theoretical analysis and controlled experiments, we show that several state-of-the-art strided convolutional encoders impose two structural bottlenecks, both predictable from architecture and signal structure, on access to these primitives: (1) they collapse primitives into alias equivalence classes, establishing a bound on representational capacity, and (2) they limit the frequency resolution available to learned filters, restricting separability. For well structured data, we find collapse rates of 31-35% and filter bandwidths 10-35x above the theoretical resolution bound, confirming that both bottlenecks arise under realistic signal conditions. We then introduce Gabor Latent Refactorization (GLRF), a lightweight post-hoc intervention that re-expresses encoder latents in a frequency-localized basis, reducing filter bandwidths from 10-35x to 1.5-3x of the theoretical resolution bound while preserving reconstruction fidelity and improving control over attributes like pitch. These results show that the encoders in question predictably degrade access to frequency-localized primitives, entangling the features that depend on them, and that a lightweight, retraining-free intervention can recover much of that access, improving steerability and interpretability.

Deep neural networks are widely deployed in high-stakes visual applications where interpretability is critical, yet existing explanations face a trade-off: post-hoc concept methods recover factors that are faithful to a model's behavior but unnamed, while naming and by-design methods attach human-readable concepts only by retraining or altering the classifier. We propose Language-Anchored Decomposition (LAD), a post-hoc framework that delivers concepts which are simultaneously named, faithful, and obtained without modifying the model. For each class, a large language model proposes a concept vocabulary that CLIP-based similarity maps localize across image regions. Inverting standard non-negative matrix factorization, LAD fixes these language-grounded maps as the coefficient matrix and learns only a concept basis that reconstructs the frozen encoder's activations, so naming becomes a structural constraint and the model's own feature geometry determines which concepts are retained. Removing this anchor preserves accuracy but collapses attribution faithfulness. Across natural-image, scene, and medical-imaging benchmarks, LAD produces spatially precise explanations that are decision-relevant under both concept insertion and deletion, while uniquely providing stable, human-interpretable concept names.

This study investigates the presence and propagation of bias within Neural Networks through a comprehensive multi-level analysis spanning the learned latent space, layer activations, and the network's parameters. Based on this taxonomy, we propose three bias detection approaches: 1) SpaceBias (new method), which characterizes the latent space prior to the final classification layer using neighbor-probability distributions and quantifies bias with the two-sample Kolmogorov-Smirnov test on the per-group distributions. 2) ActivationBias (extension of the existing method InsideBias), which analyzes the activations of neural network filters and quantifies bias via a Mann-Whitney U test, based on the observed fact that underrepresented groups exhibit lower activation levels in the final convolutional layers. 3) WeightBias (extension of the existing method IFBiD), which uses a secondary neural network trained to identify biased patterns directly in the parameters of task-specific models. Unlike conventional methods, which assess neural network outcomes and treat the model as a black box, our proposed techniques provide insight into how biases manifest within the network architecture itself at different levels, offering a more nuanced and detailed understanding. Experiments are conducted on two complementary applications: gender classification in the DiveFace dataset (72,000 face images) and digit classification on a colored-MNIST benchmark with controlled bias severity. In total, more than 127,000 models with varying degrees and types of bias were trained and evaluated. The severity sweep shows that the internal disparity, and with it the detection performance, decreases smoothly as the training distribution approaches balance. The results highlight the importance of methods that provide deeper insight into the behavior of AI models.

Lattice field theory is the workhorse of non-perturbative physics, used to simulate phenomena from the strong nuclear force to critical phenomena in materials. Its Boltzmann distributions are parametrized analytically by coupling constants, but these bare parameters are weak predictors of observables -- extracting physics typically requires extensive simulation. While normalizing flows have emerged as effective samplers at fixed couplings, it remains difficult to interpret what these networks have learned. This raises a natural question: can the physics be read off directly from the flow network parameters themselves, and can those parameters be generated for unseen theories? We propose lattice field theory as a testbed for neural network interpretability: because the target physics is qualitatively well-understood and smoothly varying, it provides ideal synthetic data with known ground truth. To this end, we introduce JEPAWG, a Joint-Embedding Predictive Architecture-based Weight Generator that maps couplings directly to flow weights via a learned latent space. On a scalar theory at lattices of size $6^2$ to $11^2$, the JEPAWG latent space recovers the correct intrinsic dimension of the underlying manifold, locates the phase transition, and encodes a finite-size shift aligned with the 2D Ising exponent $ν\approx 1$, allowing us to uncover physical structure by studying the network weights alone. This suggests the fascinating idea of treating the network weights as a new type of physical observable. As a generator, JEPAWG also interpolates and extrapolates to unseen couplings effectively and remains robust to weight-space discontinuities introduced by multi-seed training data, outperforming PCA, AE, and VAE baselines.

The architecture of deep feedforward neural networks is ubiquitous in deep learning, either as a whole system or as a subnetwork of other architectures, and thus its mechanism is a key ingredient of the black box of neural networks. On the basis of the simplest two-layer ReLU network, this paper systematically studies the mechanism of deep feedforward ReLU networks with multiple hidden layers and successfully explains the training solution obtained by the back-propagation algorithm. The concept of a path, especially in terms of the relationships between paths, plays a central role in uncovering the mystery of the black box. It is shown that a unit of a deep ReLU network can form a piecewise linear manifold to divide the input space, instead of a hyperplane of the two-layer case. How to efficiently use the hidden-layer units to produce both linear functions and partitions of the input space is also a central problem. The principles of a two-layer ReLU network can be generalized to the deeper case to a large extent, such as multiple strict partial orders and continuity restriction. The combination of the basic and simple principles proposed can yield complicated instantiations including the training solutions, and in this sense the black box of deep feedforward ReLU networks is revealed.

Foundation Models for Electronic Health Records (FEMRs) are pretrained on large-scale structured patient data, enabling them to convert longitudinal patient trajectories into generalizable representations for diverse clinical prediction tasks. Despite their effectiveness, FEMRs remain black-box models, raising concerns about bias, interpretability, and clinical trust. To address this, we propose the first token-level explainability approach for FEMRs. We train a Transformer-based surrogate model on input-output pairs from the FEMR across two prediction tasks, approximating its behavior while preserving temporal dynamics. We identify the most influential tokens, providing insights into how FEMRs leverage different aspects of patient history for predictions. To evaluate clinical relevance, we introduce a novel clinical alignment metric that quantifies the correspondence between the surrogate model's key tokens and clinically validated features. Our results demonstrate that the surrogate closely approximates FEMR predictions and that token-level explanations align well with clinical knowledge, offering a practical framework for interpretable and trustworthy clinical AI.

Neural autoregressive solvers for the Multi-Attribute Vehicle Routing Problem (MAVRP) reach competitive cost but offer no per-step justification, a problem when dispatchers must validate, accept, or compare them. We open two complementary black boxes in one protocol. On the encoder side, linear probes, spontaneous-organization metrics, rank-based richness measures, and discovered-direction analyses with intervention validation characterize how the latent represents constraint families at the graph, node, and edge level. On the decoder side, three attribution methods (gradient, integrated gradients, DeepLIFT) feed three reading angles: abductive, contrastive against the best feasible alternative, and counterfactual (smallest input change that switches the action or restores feasibility). Explanations are scored on fidelity, concentration, stability, sanity, and actionability. Across six variants combining three encoders (Attention baseline, Unimp, UnimpMoe) with two decoders (Hard-Mask, Recourse), we find that graph inductive bias improves both representational predictability and decoder sanity, that the Mixture-of-Experts encoder represents constraints in a distributed rather than axis-aligned way, and that the Recourse training regime, not merely its softer mask, produces policies that represent infeasibility usefully, exposing make-feasible counterfactuals that Hard-Mask policies fail to produce even when fed infeasible alternatives externally.

A central goal of mechanistic interpretability is to understand how neural networks work and what each individual component does. Dominant circuit-finding approaches focus on a specific behavior and reverse-engineer the role of components on the associated sub-distribution. However, past work has shown that components can have different functions that are active on different subsets of the input distribution. In this work we ask whether a single weight can be understood globally across the full training distribution by characterizing when it matters (the inputs on which ablating it changes the model's predictions). We introduce an automated LLM pipeline that writes a short, human-readable description of when a weight matters and verifies it on held-out text, crediting a weight only if its description generalizes. Across two sparse and two dense transformers, the fraction of weights that are interpretable (in this sense) is higher in sparse transformers than in dense ones, a gap that widens once unreliable descriptions are discarded. Our results show that a meaningful fraction of a sparse transformer model's weights can be interpreted: 12 to 31% of weights have a single short description that identifies what the weight is used for.

Operator learning has emerged as a powerful tool for modeling complex physical systems in functional spaces. However, their neural network-based architectures make them opaque models, obscuring the reasoning behind their predictions. In this work, we introduce a self-explainable operator learning framework that overcomes this challenge by reformulating operator learning as a linear combination of generalized functional linear models expressed through integral equations. Exploiting the additive decomposability of these integral equations, we divide the input domain into subdomains and compute localized integrals to evaluate the contribution of each region to the final prediction. This decomposition enables direct interpretability where the model explains both inputs and outputs by linking specific input regions to corresponding output patterns, thereby revealing which spatial features drive predictions. We demonstrate the framework on function-to-scalar and function-to-function mappings in fluid flow problems involving blood flow and unsteady aerodynamics. The results show that the operator most often prioritizes regions with strong feature gradients, providing physically meaningful insight into the model's decision-making process. Comparisons with established post-hoc explainability methods demonstrate qualitative agreement while highlighting the key advantage of the proposed approach: explainability is embedded directly within the operator structure itself and does not require an external tool. Therefore, our framework provides a mathematically transparent and physically interpretable approach to uncover relationships within data, fostering trust in machine learning for scientific applications by enabling more informed data-driven analysis of physical systems.

We propose a novel hybrid neural architecture, the Geometry-aware R-Structured Kolmogorov-Arnold Network (GRS-KAN), which integrates V.L.Rvachev's R-functions into the Kolmogorov-Arnold Network (KAN) framework. The proposed approach combines two complementary modeling mechanisms: smooth nonlinear structure is learned by KAN branches, while known geometric or logical constraints are encoded analytically using differentiable R-functions. This enables explicit representation of discontinuities, feasible regions, and implicit geometric boundaries within a trainable neural architecture. The framework implements differentiable logical operations through R-conjunctions and R-disjunctions, allowing complex geometric supports to be represented analytically and incorporated directly into regression models. Several GRS-KAN variants are introduced, including additive, multiplicative, and agnostic branch-weighted architectures. The method is demonstrated on regression problems involving discontinuities with circular and rectangular supports. Numerical experiments show that explicit geometric encoding substantially improves predictive accuracy and boundary localization compared with standard KANs. In the considered benchmarks, geometry-aware GRS-KAN models reduce test RMSE by up to 67% while simultaneously improving interpretability through explicit analytical representation of the learned geometric structure. The agnostic variant further demonstrates the ability to automatically determine whether geometric priors are beneficial for a given learning task.

We introduce the Neuro-Bayesian-Symbolic Residual Attention Shallow Network (NBS-RASN), a hybrid neural architecture for explainable cybersecurity risk assessment in open-source ecosystems. Unlike deep models that trade interpretability for accuracy, our shallow network encodes domain knowledge, causal reasoning, and expert judgment as differentiable components. It uses 80 interpretable neurons across 12 layers, including a gatekeeper that enforces five epistemological axioms - precision, causality, falsifiability, transparency, and completeness - as hard constraints before propagation. Despite limited depth, the network exhibits deep-learning traits via residual attention and feedback loops, learning complex risk patterns without becoming a black box. It produces fully decomposable scores: a deterministic weighted component plus an expert adjustment, with each adjustment traceable to named amplifiers (blast radius, propagation speed, structural nature, default exposure, exploitation pattern, institutional criticality). We validate on 20 open-source projects covering all OWASP Top 10:2025 categories and language risk classes, achieving confidence scores of 0.79-0.97, and show that explainability is guaranteed by design, not by a training algorithm. This challenges the assumption that deep learning requires deep networks, proving that shallow networks with deep reasoning can outperform opaque models in high-stakes cybersecurity, where interpretability is essential.

Interpretable Mesomorphic Neural Networks (IMNs) offer a promising framework that combines the predictive power of deep neural networks with the interpretability of linear models. However, the original formulation lacks safeguards to ensure that the learned interpretations are in fact reliable. In particular, the network is free to concentrate all explanatory variance into a single weight of the linear output layer, achieving strong predictive performance while producing interpretations that are largely meaningless. Paradoxically, the L1 penalty proposed to encourage sparse solutions exacerbates this problem by further incentivizing such degenerate configurations. To address this vulnerability, we introduce Local Fidelity Regularization (LFR), a novel penalty term that prevents degenerate weight collapse by aligning the linear output weights with local data variations. This structural constraint guarantees faithful explanations and substantially improves the reliability of model interpretations. Furthermore, empirical evaluations across the OpenML benchmark suite demonstrate that LFR does not compromise accuracy for explainability; rather, it achieved improved AUROC over the unregularized IMN. By yielding results highly competitive with state-of-the-art black-box models, LFR provides the dual benefit of reliable interpretability and superior predictive performance. Source code and usage instructions are available at https://github.com/hugohammer/LFR-IMN.git.

Sparse autoencoders (SAEs) have become a leading tool for interpreting the representations of vision foundation models, decomposing their polysemantic activations into a larger set of sparse, more monosemantic features. The Top-$k$ SAE, a now-standard variant, enforces sparsity architecturally through its activation function, retaining only the $k$ most active latents per input. Because it was designed precisely to avoid the $\ell_1$ penalty used by earlier SAEs and its known drawbacks, it has not been combined with an explicit sparsity regularizer. Yet the Top-$k$ SAE retains limitations of its own, and we hypothesize that a sparsity penalty acting before the selection could sharpen each latent's selectivity and make the code more interpretable, without reintroducing the drawbacks of the $\ell_1$ penalty. We introduce two sparsity regularizers compatible with the Top-$k$ architecture, both acting on the activations before the Top-$k$ selection: an $\ell_1$ penalty on the unselected (off-support) units, and a scale-invariant $\ell_1/\ell_2$-ratio penalty that concentrates the code onto fewer effective units. Both penalties are applied only to the batch-active units, those selected by the Top-$k$ operator at least once within the batch. Across two datasets, three vision foundation models, and a range of $k$, both regularizers consistently improve monosemanticity at no cost to reconstruction quality. The $\ell_1/\ell_2$ penalty further concentrates information into fewer latents, making reconstruction more robust to the inference-time choice of $k$ and improving small-budget linear probing. Our central finding is that hard architectural sparsity and soft sparsity regularization are complementary rather than mutually exclusive.

Given the widespread prevalence of depression and its consequential impact on individuals and society, it is crucial to obtain objective measures for early diagnosis and intervention. As a multidisciplinary topic, these objective measures should be interpretable and accessible to health care professionals, ensuring effective collaboration and treatment planning in the realm of mental health care. Even though current automated depression diagnosis approaches improved over the last decade, a critical gap exists as they often lack affect-specificity and interpretability, limiting their practical application and potential impact on mental health care. In particular, interpretability from temporal activities from videos when deep models are used is not fully explored. In this study, we present a novel framework for analyzing Deep Neural Networks' decisions when trained on facial videos, specifically focusing on automatic depression severity diagnosis. By fine-tuning Deep Convolutional Neural Networks (DCNN) pre-trained on Action Recognition datasets on depression severity facial videos from AVEC depression dataset, our framework is able to interpret the model's saliency maps by examining face regions and temporal expression semantics. Our approach generates both visual and quantitative explanations for the model's decisions, providing greater insight into its reasoning. In addition to this interpretability, our video-based modeling has improved upon previous single-face benchmarks for visual depression diagnosis, resulting in enhanced predictive performance. Overall, our work demonstrates the successful development of a framework capable of generating hypotheses from a facial model's decisions while simultaneously improving depression's predictive capabilities.

What is the geometry of a visual percept? The most widely used protocols for decomposing neural network representations into interpretable parts treat concepts as isolated directions, yet recent work shows that concepts are often realized as geometric structures in low dimensional regions of activation space. We turn to the literature of Structured sparsity to close this gap, and show that block sparsity, which groups directions into blocks, is the prior matched to a generative model in which a representation is a sparse sum of low-dimensional manifolds: the modern, learned form of a classical idea in visual neuroscience, where a visual feature is carried by a coordinated group of neurons rather than a single tuned one. We implement three variants of block-sparse featurizers (BSFs) and, through a minimum-description-length analysis, show that all three describe activations more compactly than direction-based featurizers, with the recovered concepts typically two- to four-dimensional. We then use BSFs to (i) recontextualize prior work, showing that curve detectors in InceptionV1 actually read from a single continuous curve manifold, (ii) discover novel manifolds including shadows and lighting in DINOv3, and (iii) support interpretable control of image generation in diffusion models (SDXL) via manifold steering.

Accurate electricity load forecasting is a crucial prerequisite for stable power system operations. While prevalent deep learning models present competitive performance, they often operate as black boxes and lack interpretability. While the Kolmogorov-Arnold network (KAN) has emerged as a promising alternative because of its learnable activation function design, its direct application to time-series forecasting faces challenges in modeling complex temporal data patterns. Also, simple integration into existing architectures, such as serving as replacement of neural modules, cannot fully leverage KAN's interpretability strengths. To address these gaps, this study develops LoadKAN, a novel hybrid and interpretable framework for load forecasting that synergistically combines a specifically-designed feature-isolated temporal attention mechanism with a KAN module. The attention stage aims to extract temporal dynamics from each input feature independently, such as historical load and human mobility, providing distilled feature representations to the KAN module for interpretable predictions. When evaluated on datasets from three representative U.S. electricity markets, our LoadKAN remains highly competitive when compared to extensively-tuned, state-of-the-art, black-box deep learning benchmarks. More importantly, LoadKAN's interpretability enables a granular analysis of the learned non-linear relationships between six distinct mobility patterns and electricity load. Through KAN-learned activation functions, our quantitative sensitivity analyses on mobility features reveal complex and market-specific dependencies. These findings further demonstrate the ability of our LoadKAN to generate insights often obscured by opaque black-box neural forecasting models.

Explaining and understanding the decision-making process of artificial intelligence (AI) systems, particularly those implemented by neural networks, falls within the field of explainable AI (XAI). Analogous to the human attention mechanism, neural networks are assumed to possess their own attention mechanisms that selectively process information during decision-making. This work proposes to study one XAI topic: analysing and visualising the attention mechanisms of neural networks. Our experiments are performed on speaker recognition neural networks that are trained to identify speaker identity from a given utterance. Previous studies have widely used class activation map (CAM)-based methods to analyse and visualise the attention mechanisms of neural networks. Each of these methods produces an attention map for each network input, highlighting which input regions are selectively processed when the speaker recognition network makes decisions. However, the evaluation of attention maps produced by these methods remains largely underexplored. This work systematically reviews an existing attention map evaluation algorithm, establishing key concepts and identifying its shortcomings. On the basis of this existing evaluation algorithm, a new version is then proposed to address the identified shortcomings, called the Modified Randomised Input Sampling for Explanation - Evaluation algorithm (Modified RISE-eval). Using Modified RISE-eval, we evaluate the attention maps produced by two representative CAM-based methods, GradCAM and LayerCAM, applied to a certain speaker recognition network. The evaluation results demonstrate that GradCAM and LayerCAM each exhibit distinct advantages when applied under different experimental conditions in the speaker recognition task.

Electroencephalogram (EEG) analysis remains the clinical gold standard for epilepsy diagnosis and seizure detection. While Deep Learning (DL) has significantly advanced automated EEG interpretation, its transition from controlled experimental settings to routine clinical deployment is severely bottlenecked by fundamental architectural flaws. Standard DL models operate as opaque black-boxes lacking clinical interpretability, demand massive amounts of balanced annotated data, and incur steep computational costs incompatible with resource-constrained wearable or implantable neuromodulation devices. This paper presents a comprehensive review of these prevailing limitations and explores Kolmogorov-Arnold Networks (KANs) as a emerging paradigm for EEG-based seizure detection. By replacing the fixed activation functions of traditional neurons with flexible, learnable functions along the network's connections, KANs bridge the critical gap between predictive accuracy and mathematical transparency. We systematically analyze how KAN architectures resolve the shortcomings of traditional DL-based models by offering exceptional parameter efficiency, inherent interpretability for physician trust, and robust performance under data scarcity. Ultimately, this review establishes KANs not merely as an incremental algorithmic update, but as a fundamental paradigm shift necessary to actualize next-generation, patient-specific, and thoroughly transparent clinical EEG monitoring systems.

Neural networks outperform classical GCC-PHAT for Time-Difference-of-Arrival (TDOA) estimation in noise and reverberation, yet their internal strategy remains unexplored. To uncover it, we turn GCC-PHAT's mathematical steps into diagnostic targets, probing hidden layers of three architectures (MLP, CNN, Transformer) and complementing with gradient attribution and causal frequency masking. We find that cross-power computation consistently emerges across all architectures and conditions, while PHAT whitening, the defining step of GCC-PHAT, fails to emerge. Instead, networks learn a magnitude-aware frequency weighting that preserves per-frequency reliability information discarded by PHAT. This makes PHAT an information bottleneck: removing it from both classical and neural GCC pipelines improves performance under additive noise. On real-world reverberant data, PHAT remains the best classical weighting, but end-to-end networks achieve lower error by learning data-adaptive weighting.

We study what Lorentz-equivariant jet taggers learn internally, using equivariance tests, linear probes and grade ablations across five models including L-GATr, L-GATr-slim and LLoCa-T. Linear probes show that equivariant models suppress frame-dependent pseudorapidity to zero while encoding jet mass and N-subjettiness strongly. Grade ablations on L-GATr reveal that bivector channels are negligible for top-quark tagging while vector-like channels are dominant but seed variable, consistent with the network exploiting multiple representational pathways. These results characterize which physical features and algebraic grade structures carry discriminative information in equivariant taggers and may inform future development of such models.

Parameter decomposition (PD) decomposes neural networks into interpretable computational components that faithfully reflect the original network's operations. However, scaling PD to large models requires vast compute, making it a costly and risky endeavor. Here we propose targeted PD (tPD), which identifies only the components that process specific inputs of interest -- from isolated prompts to large subtasks -- by introducing a high-rank catch-all component that handles all non-target data. We validate tPD on toy models and on transformer language models trained on The Pile, where it recovers reproducible, mechanistically faithful circuits. We extract a CSS-only submodel of a 4-block transformer using 7% of the FLOPs of its published decomposition, and in a 12-block transformer we surgically ablate and rewire memorized sequences, with negligible side effects on other inputs.

Deep neural network-based automatic speaker verification (ASV) systems achieve impressive performance but their embedding representations remain opaque, lacking a structured and perceptually verifiable explanation of the vocal characteristics they encode. Existing approaches either require annotation of speaker attributes or introduce alternative representations whose interpretability is unvalidated with listeners. We propose Listenable Interpretable Speaker Embeddings (LISE), a label-free framework that decomposes pretrained speaker embeddings into a small set of components. This decomposition yields a structured representation that supports the analysis of what information has been encoded by speaker embeddings. LISE preserves ASV performance with negligible EER degradation on x-vector and ECAPA-TDNN. Crucially, the interpretability of these components for human listeners is demonstrated through listening experiments, where participants distinguished speakers with 83.9% accuracy.

Video-based seizure detection is essential for the management of epilepsy patients, offering a non-invasive complement to electroencephalography. While several deep learning approaches have been developed for video-based seizure detection, none are inherently interpretable, limiting their adoption and translation into clinical practice. We present, to our knowledge, the first exploration of a neurosymbolic framework for video-based seizure detection that directly addresses this gap. Our approach (1) extracts patient-centric skeleton sequences from epilepsy monitoring units via a prompt-guided foundation model, (2) predicts binary spatio-temporal concept activations grounded in clinical motor semiology guidelines, and (3) composes them via differentiable logic into interpretable Boolean rules with auditable contributions. Furthermore, to mitigate false positives arising from the traditional binary formulation (seizure vs.\ non-seizure), we sub-classify non-seizure segments into clinically relevant normal activities, providing the model with fine-grained discriminative supervision. Evaluated on two public seizure video benchmarks, our framework achieves 89.78% sensitivity with 0.06 false detections per hour on SAHZU and 85.27%,0.09 on IEEE, while producing complete three-level interpretability: every prediction decomposes into which motor primitives were detected, how they were logically composed, and how much each rule contributed to the clinical decision. We publicly release all annotations, extracted pose sequences, our data pipeline and code, https://github.com/Mr-TalhaIlyas/CDSD/.

Neural networks learn features that reflect the hierarchical, multi-scale structure of natural data. Synthetic datasets used to evaluate interpretability methods typically lack this structure, limiting their value as realistic toy models. To close this gap, we introduce a family of synthetic datasets consisting of hierarchical functions defined on critical mean-field percolation clusters embedded in a high-dimensional data space. The percolation data consists of sparse, low-dimensional fractal clusters with a power-law size distribution. Latent variables modeling a taxonomic hierarchy generate each data point's target value. The data model is analytically tractable with known critical exponents that fix its properties without requiring hyperparameter tuning. We leverage a mapping between percolation clusters, random trees, and additive coalescence to propose an almost linear-time algorithm to jointly sample a random tree and its hierarchical latent decomposition, enabling data generation at arbitrary scale. Using probing experiments, we find that the model's ground-truth latent variables can be linearly decoded from neural network activations. Together, sparsity, self-similarity, power-law statistics, and analytical tractability make critical percolation a principled testbed for interpretability research.

Deep neural networks (DNNs) exhibit attractive performance in various fields but often suffer from low interpretability. The neural additive model (NAM) and its variant called the neural basis model (NBM) use neural networks (NNs) as nonlinear shape functions in generalized additive models (GAMs). Both models are highly interpretable and exhibit good performance and flexibility for NN training. NAM and NBM can provide and visualize the contribution of each feature to the prediction owing to GAM-based architectures. However, when using two-input NNs to consider feature interactions or when applying them to high-dimensional datasets, training NAM and NBM becomes intractable due to the increase in the computational resources required. This paper proposes incorporating the feature selection mechanism into NAM and NBM to resolve computational bottlenecks. We introduce the feature selection layer in both models and update the selection weights during training. Our method is simple and can reduce computational costs and model sizes compared to vanilla NAM and NBM. In addition, it enables us to use two-input NNs even in high-dimensional datasets and capture feature interactions. We demonstrate that the proposed models are computationally efficient compared to vanilla NAM and NBM, and they exhibit better or comparable performance with state-of-the-art GAMs.

Neural Combinatorial Optimization (NCO) achieves strong performance, yet its black-box nature remains a key roadblock to deployment and scientific diagnosis. Standard interpretability tools, such as Concept Bottleneck Models (CBMs), are ill-equipped for NCO, whose decisions are dynamic, state-dependent, and lack proper concept vocabulary definition. To close this gap, we introduce Evolving Programmatic Bottlenecks (EPB), to our knowledge, the first framework for interpreting NCO policies by distilling black-box NCO models into human-readable program portfolios. EPB employs an LLM to autonomously evolve a bank of programs, where each program's per-step action distribution serves as the bottleneck. EPB works through an iterative framework: Block I fixes program bank capacity and introduces a hybrid textual-numerical gradient descent scheme that couples numerical gradients for student router updates and textual gradients for LLM-based program revision; Block II dynamically adapts bank capacity via fault-targeted expansion and redundancy pruning. Extensive experiments demonstrate EPB's effectiveness and broad applicability, where the distilled program portfolios largely match original performance. EPB also reveals that NCO behavior shifts across optimization stages and can be approximated as a composition of classic heuristic variants. Our work advances interpretable NCO and establishes EPB as a promising tool for interpreting sequential decision-making models.