new

Get trending papers in your email inbox!

Subscribe

Daily Papers

byAK and the research community

Nov 19

ABLE: Representing and Mapping LLMs via Attribution-Based Large-model Embedding

The explosive growth of large language models (LLMs) has created a heterogeneous and poorly documented ecosystem, making systematic model comparison increasingly important for provenance auditing, security analysis, and model selection. Existing representation methods struggle to address this setting efficiently. Approaches analyzing internal parameters are powerful when architectures are compatible, but face scalability barriers under structural heterogeneity, while methods relying on external outputs may conflate models with similar behaviors and are difficult to align in richer output spaces across different tokenizers. To bridge this gap, we propose ABLE (Attribution-Based Large-model Embedding), a framework that leverages the interpretability space to construct model representations. By aggregating gradient-based feature attributions via a tokenizer-agnostic word-level alignment, ABLE captures model-specific input-sensitivity patterns rather than only surface-level outputs. Beyond empirical utility, we provide a stability analysis showing that, under standard regularity assumptions for differentiable Transformer-style models, ABLE induces a Lipschitz-continuous parameter-to-embedding map with finite-sample convergence guarantees. Extensive experiments on 239 open-source LLMs demonstrate that our training-free approach achieves competitive or superior performance in relation prediction, model routing, and benchmark score prediction.

  • 7 authors
·
Aug 28

SESA: Supervised Explicit Semantic Analysis

In recent years supervised representation learning has provided state of the art or close to the state of the art results in semantic analysis tasks including ranking and information retrieval. The core idea is to learn how to embed items into a latent space such that they optimize a supervised objective in that latent space. The dimensions of the latent space have no clear semantics, and this reduces the interpretability of the system. For example, in personalization models, it is hard to explain why a particular item is ranked high for a given user profile. We propose a novel model of representation learning called Supervised Explicit Semantic Analysis (SESA) that is trained in a supervised fashion to embed items to a set of dimensions with explicit semantics. The model learns to compare two objects by representing them in this explicit space, where each dimension corresponds to a concept from a knowledge base. This work extends Explicit Semantic Analysis (ESA) with a supervised model for ranking problems. We apply this model to the task of Job-Profile relevance in LinkedIn in which a set of skills defines our explicit dimensions of the space. Every profile and job are encoded to this set of skills their similarity is calculated in this space. We use RNNs to embed text input into this space. In addition to interpretability, our model makes use of the web-scale collaborative skills data that is provided by users for each LinkedIn profile. Our model provides state of the art result while it remains interpretable.

  • 2 authors
·
Aug 10, 2017

U-Space: Uncovering When and Why Uncertainty Arises in Language Models

Large language models are informing decisions with ever-higher stakes. As the consequences of their errors grow, a central question becomes harder to ignore: how much can we trust an individual answer? Yet recognizing when to defer remains difficult because language models can present incorrect conclusions with fluent explanations and an authoritative tone. Uncertainty quantification seeks to address this disconnect by estimating the reliability of individual predictions. However, many existing methods require repeated generations or separately trained components, and their scalar estimates do not reveal where uncertainty arises or how it evolves during reasoning. Recent work has also shown that generation length can be strongly associated with uncertainty estimates and correctness, raising the question of how much of an estimator's predictive power comes from uncertainty-specific information rather than output length alone. Mechanistic interpretability offers a way to address these limitations by connecting human-interpretable concepts to intermediate model states. Building on this capability, we introduce the U-Space, a low-dimensional subspace that makes a model's evolving uncertainty measurable and interpretable. We identify semantic anchors for doubt and certainty, map their unembedding directions back into the residual space, and combine their contrasts into an orthogonal basis. The U-Lens projects each token state onto these basis vectors, yielding an interpretable token-level uncertainty map that can be inspected directly or aggregated into a scalar uncertainty score. Our approach requires no correctness labels, repeated generations, or training. Across reasoning benchmarks, its confidence score outperforms established baselines under both standard and length-controlled evaluation and transfers more reliably than supervised estimators. Code: https://github.com/s2labres/U-Space.

Diagnosing Transformers: Illuminating Feature Spaces for Clinical Decision-Making

Pre-trained transformers are often fine-tuned to aid clinical decision-making using limited clinical notes. Model interpretability is crucial, especially in high-stakes domains like medicine, to establish trust and ensure safety, which requires human engagement. We introduce SUFO, a systematic framework that enhances interpretability of fine-tuned transformer feature spaces. SUFO utilizes a range of analytic and visualization techniques, including Supervised probing, Unsupervised similarity analysis, Feature dynamics, and Outlier analysis to address key questions about model trust and interpretability. We conduct a case study investigating the impact of pre-training data where we focus on real-world pathology classification tasks, and validate our findings on MedNLI. We evaluate five 110M-sized pre-trained transformer models, categorized into general-domain (BERT, TNLR), mixed-domain (BioBERT, Clinical BioBERT), and domain-specific (PubMedBERT) groups. Our SUFO analyses reveal that: (1) while PubMedBERT, the domain-specific model, contains valuable information for fine-tuning, it can overfit to minority classes when class imbalances exist. In contrast, mixed-domain models exhibit greater resistance to overfitting, suggesting potential improvements in domain-specific model robustness; (2) in-domain pre-training accelerates feature disambiguation during fine-tuning; and (3) feature spaces undergo significant sparsification during this process, enabling clinicians to identify common outlier modes among fine-tuned models as demonstrated in this paper. These findings showcase the utility of SUFO in enhancing trust and safety when using transformers in medicine, and we believe SUFO can aid practitioners in evaluating fine-tuned language models for other applications in medicine and in more critical domains.

  • 6 authors
·
May 27, 2023

Foundation Models for Discovery and Exploration in Chemical Space

Accurate prediction of atomistic, thermodynamic, and kinetic properties from molecular structures underpins materials innovation. Existing computational and experimental approaches lack the scalability required to efficiently navigate chemical space. Scientific foundation models trained on large unlabeled datasets offer a path toward exploring chemical space across diverse application domains. Here we develop MIST, a family of molecular foundation models with up to an order of magnitude more parameters and data than prior works. Trained using a novel tokenization scheme that comprehensively captures nuclear, electronic, and geometric information, MIST learns from a diverse range of molecules. MIST models have been fine-tuned to predict more than 400 structure -- property relationships and match or exceed state-of-the-art performance across benchmarks spanning physiology, electrochemistry, and quantum chemistry. We demonstrate the ability of these models to solve real-world problems across chemical space, including multiobjective electrolyte solvent screening, olfactory perception mapping, isotope half-life prediction, stereochemical reasoning for chiral organometallic compounds, and binary and multi-component mixture property prediction. Probing MIST models using mechanistic interpretability methods reveals identifiable patterns and trends not explicitly present in the training data, suggesting that the models learn generalizable scientific concepts. We formulate hyperparameter-penalized Bayesian neural scaling laws and use them to reduce the computational cost of model development by an order of magnitude. The methods and findings presented here represent a significant step toward accelerating materials discovery, design, and optimization using foundation models and provide valuable guidance for training compute-optimal scientific foundation models.

  • 22 authors
·
Oct 20, 2025

Exemplar Partitioning for Mechanistic Interpretability

We introduce Exemplar Partitioning (EP), an unsupervised method for constructing interpretable feature dictionaries from large language model activations with sim 10^3times fewer tokens than comparable sparse autoencoders (SAEs). An EP dictionary is a Voronoi partition of activation space, built by leader-clustering streamed activations within a distance threshold. Each region is anchored by an observed exemplar that serves as both its membership criterion and intervention direction; dictionary size is not prespecified, but determined by the activation geometry at that threshold. Because exemplars are observed rather than learned, dictionaries built from the same data stream are directly comparable across layers, models, and training checkpoints. We characterise EP as an interpretability object via targeted demonstrations of properties newly accessible through this construction, plus one head-to-head benchmark. In Gemma-2-2B, EP dictionary regions are interpretable and support causal interventions: refusal in instruction-tuned Gemma concentrates in a region whose exemplar ablation can collapse held-out refusal. Cross-checkpoint matching between base and instruction-tuned dictionaries separates the directions preserved through finetuning from those introduced by it. EP regions and Gemma Scope SAE features decompose activation space differently but agree on a shared core: sim20% of EP regions match an SAE feature at F_1 > 0.5, and EP one-hot probes retain sim97% of raw-activation probe accuracy at ell_0 = 1. Nearest-exemplar distance provides a free out-of-distribution signal at inference. On AxBench latent concept detection at Gemma-2-2B-it L20, EP at p_1 reaches mean AUROC 0.881, +0.126 over the canonical GemmaScope SAE leaderboard entry and within 0.030 of SAE-A's 0.911, at sim 10^3times less build compute.

  • 1 authors
·
May 14

On Interpretability of Deep Learning based Skin Lesion Classifiers using Concept Activation Vectors

Deep learning based medical image classifiers have shown remarkable prowess in various application areas like ophthalmology, dermatology, pathology, and radiology. However, the acceptance of these Computer-Aided Diagnosis (CAD) systems in real clinical setups is severely limited primarily because their decision-making process remains largely obscure. This work aims at elucidating a deep learning based medical image classifier by verifying that the model learns and utilizes similar disease-related concepts as described and employed by dermatologists. We used a well-trained and high performing neural network developed by REasoning for COmplex Data (RECOD) Lab for classification of three skin tumours, i.e. Melanocytic Naevi, Melanoma and Seborrheic Keratosis and performed a detailed analysis on its latent space. Two well established and publicly available skin disease datasets, PH2 and derm7pt, are used for experimentation. Human understandable concepts are mapped to RECOD image classification model with the help of Concept Activation Vectors (CAVs), introducing a novel training and significance testing paradigm for CAVs. Our results on an independent evaluation set clearly shows that the classifier learns and encodes human understandable concepts in its latent representation. Additionally, TCAV scores (Testing with CAVs) suggest that the neural network indeed makes use of disease-related concepts in the correct way when making predictions. We anticipate that this work can not only increase confidence of medical practitioners on CAD but also serve as a stepping stone for further development of CAV-based neural network interpretation methods.

  • 6 authors
·
May 5, 2020

NSTR: Neural Spectral Transport Representation for Space-Varying Frequency Fields

Implicit Neural Representations (INRs) have emerged as a powerful paradigm for representing signals such as images, audio, and 3D scenes. However, existing INR frameworks -- including MLPs with Fourier features, SIREN, and multiresolution hash grids -- implicitly assume a global and stationary spectral basis. This assumption is fundamentally misaligned with real-world signals whose frequency characteristics vary significantly across space, exhibiting local high-frequency textures, smooth regions, and frequency drift phenomena. We propose Neural Spectral Transport Representation (NSTR), the first INR framework that explicitly models a spatially varying local frequency field. NSTR introduces a learnable frequency transport equation, a PDE that governs how local spectral compositions evolve across space. Given a learnable local spectrum field S(x) and a frequency transport network F_θ enforcing nabla S(x) approx F_θ(x, S(x)), NSTR reconstructs signals by spatially modulating a compact set of global sinusoidal bases. This formulation enables strong local adaptivity and offers a new level of interpretability via visualizing frequency flows. Experiments on 2D image regression, audio reconstruction, and implicit 3D geometry show that NSTR achieves significantly better accuracy-parameter trade-offs than SIREN, Fourier-feature MLPs, and Instant-NGP. NSTR requires fewer global frequencies, converges faster, and naturally explains signal structure through spectral transport fields. We believe NSTR opens a new direction in INR research by introducing explicit modeling of space-varying spectrum.

  • 1 authors
·
Nov 23, 2025

Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs

Large language models appear to develop internal representations of emotion -- "emotion circuits," "emotion neurons," and structured emotional manifolds have been reported across multiple model families. But every study making these claims uses stimuli signalled by explicit emotion keywords, leaving a fundamental question unanswered: do these circuits detect genuine emotional meaning, or do they detect the word "devastated"? We present the first clinical validity test of emotion circuit claims using mechanistic interpretability methods grounded in clinical psychology -- clinical vignettes that evoke emotions through situational and behavioural cues alone, emotion keywords removed. Across six models (Llama-3.2-1B, Llama-3-8B, Gemma-2-9B; base and instruct variants), we apply four convergent mechanistic interpretability methods -- linear probing, causal activation patching, knockout experiments, and representational geometry -- and discover two dissociable emotion processing mechanisms. Affect reception -- detecting emotionally significant content -- operates with near-perfect accuracy (AUROC 1.000), consistent with early-layer saturation, and replicates across all six models. Emotion categorization -- mapping affect to specific emotion labels -- is partially keyword-dependent, dropping 1-7% without keywords and improving with scale. Causal activation patching confirms keyword-rich and keyword-free stimuli share representational space, transferring affective salience rather than emotion-category identity. These findings falsify the keyword-spotting hypothesis, establish a novel mechanistic dissociation, and introduce clinical stimulus methodology as a rigorous standard for testing emotion processing claims in large language models -- with direct implications for AI safety evaluation and alignment. All stimuli, code, and data are released for replication.

  • 1 authors
·
Mar 14

Mechanistic Interpretability of RNNs emulating Hidden Markov Models

Recurrent neural networks (RNNs) provide a powerful approach in neuroscience to infer latent dynamics in neural populations and to generate hypotheses about the neural computations underlying behavior. However, past work has focused on relatively simple, input-driven, and largely deterministic behaviors - little is known about the mechanisms that would allow RNNs to generate the richer, spontaneous, and potentially stochastic behaviors observed in natural settings. Modeling with Hidden Markov Models (HMMs) has revealed a segmentation of natural behaviors into discrete latent states with stochastic transitions between them, a type of dynamics that may appear at odds with the continuous state spaces implemented by RNNs. Here we first show that RNNs can replicate HMM emission statistics and then reverse-engineer the trained networks to uncover the mechanisms they implement. In the absence of inputs, the activity of trained RNNs collapses towards a single fixed point. When driven by stochastic input, trajectories instead exhibit noise-sustained dynamics along closed orbits. Rotation along these orbits modulates the emission probabilities and is governed by transitions between regions of slow, noise-driven dynamics connected by fast, deterministic transitions. The trained RNNs develop highly structured connectivity, with a small set of "kick neurons" initiating transitions between these regions. This mechanism emerges during training as the network shifts into a regime of stochastic resonance, enabling it to perform probabilistic computations. Analyses across multiple HMM architectures - fully connected, cyclic, and linear-chain - reveal that this solution generalizes through the modular reuse of the same dynamical motif, suggesting a compositional principle by which RNNs can emulate complex discrete latent dynamics.

  • 5 authors
·
Oct 29, 2025

Decomposing multimodal embedding spaces with group-sparse autoencoders

The Linear Representation Hypothesis asserts that the embeddings learned by neural networks can be understood as linear combinations of features corresponding to high-level concepts. Based on this ansatz, sparse autoencoders (SAEs) have recently become a popular method for decomposing embeddings into a sparse combination of linear directions, which have been shown empirically to often correspond to human-interpretable semantics. However, recent attempts to apply SAEs to multimodal embedding spaces (such as the popular CLIP embeddings for image/text data) have found that SAEs often learn "split dictionaries", where most of the learned sparse features are essentially unimodal, active only for data of a single modality. In this work, we study how to effectively adapt SAEs for the setting of multimodal embeddings while ensuring multimodal alignment. We first argue that the existence of a split dictionary decomposition on an aligned embedding space implies the existence of a non-split dictionary with improved modality alignment. Then, we propose a new SAE-based approach to multimodal embedding decomposition using cross-modal random masking and group-sparse regularization. We apply our method to popular embeddings for image/text (CLIP) and audio/text (CLAP) data and show that, compared to standard SAEs, our approach learns a more multimodal dictionary while reducing the number of dead neurons and improving feature semanticity. We finally demonstrate how this improvement in alignment of concepts between modalities can enable improvements in the interpretability and control of cross-modal tasks.

  • 3 authors
·
Jan 26

Evaluation and Improvement of Interpretability for Self-Explainable Part-Prototype Networks

Part-prototype networks (e.g., ProtoPNet, ProtoTree and ProtoPool) have attracted broad research interest for their intrinsic interpretability and comparable accuracy to non-interpretable counterparts. However, recent works find that the interpretability from prototypes is fragile, due to the semantic gap between the similarities in the feature space and that in the input space. In this work, we strive to address this challenge by making the first attempt to quantitatively and objectively evaluate the interpretability of the part-prototype networks. Specifically, we propose two evaluation metrics, termed as consistency score and stability score, to evaluate the explanation consistency across images and the explanation robustness against perturbations, respectively, both of which are essential for explanations taken into practice. Furthermore, we propose an elaborated part-prototype network with a shallow-deep feature alignment (SDFA) module and a score aggregation (SA) module to improve the interpretability of prototypes. We conduct systematical evaluation experiments and provide substantial discussions to uncover the interpretability of existing part-prototype networks. Experiments on three benchmarks across nine architectures demonstrate that our model achieves significantly superior performance to the state of the art, in both the accuracy and interpretability. Codes are available at https://github.com/hqhQAQ/EvalProtoPNet.

  • 7 authors
·
Dec 12, 2022

CodeBrain: Towards Decoupled Interpretability and Multi-Scale Architecture for EEG Foundation Model

Electroencephalography (EEG) provides real-time insights into brain activity and supports diverse applications in neuroscience. While EEG foundation models (EFMs) have emerged to address the scalability issues of task-specific models, current approaches still yield clinically uninterpretable and weakly discriminative representations, inefficiently capture global dependencies, and neglect important local neural events. We present CodeBrain, a two-stage EFM designed to fill this gap. In the first stage, we introduce the TFDual-Tokenizer, which decouples heterogeneous temporal and frequency EEG signals into discrete tokens, quadratically expanding the representation space to enhance discriminative power and offering domain-specific interpretability by suggesting potential links to neural events and spectral rhythms. In the second stage, we propose the multi-scale EEGSSM architecture, which combines structured global convolution with sliding window attention to efficiently capture both sparse long-range and local dependencies, reflecting the brain's small-world topology. Pretrained on the largest public EEG corpus, CodeBrain achieves strong generalization across 8 downstream tasks and 10 datasets under distribution shifts, supported by comprehensive ablations, scaling-law analyses, and interpretability evaluations. Both code and pretraining weights will be released in the future version.

  • 7 authors
·
Sep 24, 2025

RevengeBench: Reverse Engineering Code-Space Policies from Behavioral Experiments

For most of scientific history, researchers studying behavior could only infer hidden mechanisms from outward actions: an inverse problem that becomes more tractable when observation is augmented by targeted intervention. We pose a computational analogue: given only behavioral traces of an agent in a game environment, can a learner reconstruct the underlying decision program as executable code, and how much does this reconstruction improve with the ability to design controlled experiments? We introduce RevengeBench, a benchmark of 75 LLM generated, Elo-calibrated policies across five game environments, drawn from CodeClash tournament trajectories. The learner observes the hidden target policy play against sampled opponents and designs behavioral probes in the form of custom opponent policies that elicit informative behavior. It then submits an executable hypothesis, which is evaluated using continuous action-distance metrics. We further validate that recovered code carries informative signal in downstream player-versus-player tournaments. Across twelve frontier LLMs, recovery quality varies substantially (34 to 72% of initial distance closed), with reconstructed policies yielding measurable competitive advantage, particularly for weaker models that otherwise struggle to design effective counter-strategies. Our benchmark positions behavioral recovery of programmatic policies as a tractable inverse problem in code-space, opening a path to opponent modeling, policy interpretability, and the broader question of inferring latent mechanisms from observations.

  • 5 authors
·
Jun 23

ProLaViT: Learning Progressive Latent Visual Thoughts in Structured Latent Space

Multimodal Large Language Models (MLLMs) have achieved remarkable progress but still struggle with complex visual reasoning tasks requiring multi-step perception and logical deduction. While explicit visual generation incurs prohibitive computational costs, existing latent approaches often rely on external experts or lack rigorous cognitive logic. In this paper, we introduce ProLaViT (Progressive Latent Visual Thought), a framework empowering MLLMs to perform structured visual derivation in the continuous latent space. Unlike works dependent on heterogeneous external models, ProLaViT leverages an endogenous self-distillation mechanism, utilizing the model's own visual encoder to supervise latent thoughts. To facilitate this, we construct a scalable programmatic synthesis pipeline enabling the model to internalize algorithmic precision without inference time tools. We design two reasoning paradigms: (1) Coarse-to-Fine Causal Chain for spatial tasks, guiding attention from global context to local targets. (2) Dialectical Reasoning Chain for logical tasks, incorporating counter-factual thinking for verification. Furthermore, we propose a Distance-Weighted Diversity Loss to impose topology-aware constraints, preventing feature degeneration by enforcing semantic distinctiveness. Extensive experiments demonstrate that ProLaViT outperforms baselines on vision-centric benchmarks, achieving superior accuracy and interpretability with high efficiency.

  • 7 authors
·
Jul 2

Do Vision-Language-Action Models Understand Instructions? A Mechanistic Interpretability Study on Language Grounding

Vision-Language-Action models are designed to generalise across environments and task descriptions, raising the question of whether their action generation actually depends on the language instruction, or whether they largely rely on visual cues and superficial correlations. Robustness to variance in the visual and linguistic observation space is critical for real-world deployment, yet VLAs lack explicit grounding modules and instead rely on the intrinsic language grounding capabilities of their Vision-Language model backbones. For this reason, we conduct a controlled mechanistic interpretability study on the language grounding capabilities of two state-of-the-art Vision-Language-Action models, π_{0.5} and GR00T N1.7, by applying activation and attribution patching to the residual stream of the action generation modules. We systematically corrupt the task instruction of input samples of the LIBERO benchmark following five strategies: synonym replacement, semantic scaling, directional corruption, random object substitution, and empty string. Our experiments find that both models are comparatively insensitive to abstract rephrasing and to referencing non-existent objects, but react strongly to empty task descriptions and, especially, to directional language. During action generation, this sensitivity is concentrated in different loci for each model: mainly in the early, periodic cross-attention layers for GR00T N1.7, versus distributed across the earliest and selected later layers for π_{0.5}. For GR00T N1.7, directional perturbations drive some of the largest causal effects while leaving the internal representational geometry comparatively unchanged, a dissociation we do not observe clearly for π_{0.5}. Finally, the reliability of attribution patching is model-dependent: it closely tracks activation patching for GR00T N1.7 but not for π_{0.5}.

  • 2 authors
·
Oct 6

Detecting Multi-Agent Collusion Through Multi-Agent Interpretability

As LLM agents are increasingly deployed in multi-agent systems, they introduce risks of covert coordination that may evade standard forms of human oversight. While linear probes on model activations have shown promise for detecting deception in single-agent settings, collusion is inherently a multi-agent phenomenon, and the use of internal representations for detecting collusion between agents remains unexplored. We introduce NARCBench, a benchmark for evaluating collusion detection under environment distribution shift, and propose five probing techniques that aggregate per-agent deception scores to classify scenarios at the group level. Our probes achieve 1.00 AUROC in-distribution and 0.60--0.86 AUROC when transferred zero-shot to structurally different multi-agent scenarios and a steganographic blackjack card-counting task. We find that no single probing technique dominates across all collusion types, suggesting that different forms of collusion manifest differently in activation space. We also find preliminary evidence that this signal is localised at the token level, with the colluding agent's activations spiking specifically when processing the encoded parts of their partner's message. This work takes a step toward multi-agent interpretability: extending white-box inspection from single models to multi-agent contexts, where detection requires aggregating signals across agents. These results suggest that model internals provide a complementary signal to text-level monitoring for detecting multi-agent collusion, particularly for organisations with access to model activations. Code and data are available at https://github.com/aaronrose227/narcbench.

  • 4 authors
·
Mar 31

Prototype-Guided Classification Sub-Task Decoupling Framework: Enhancing Generalization and Interpretability for Multivariate Time Series

Time Series Classification (TSC) is a long-standing research problem that has gained increasing attention in recent years with the rapid growth of large-scale temporal data. Despite substantial progress enabled by deep learning, designing TSC models that are both accurate and interpretable remains a challenging task. Many existing approaches adopt a direct feature-to-label classification paradigm, by collapsing high-dimensional temporal embeddings into class logits via a single linear projection (often after global pooling), the paradigm conflates feature extraction and decision logic into an inseparable mapping. To address these limitations, we propose PDFTime, a prototype-guided framework that reformulates time series classification as a multi-stage decision process. Instead of direct feature-to-label mapping, PDFTime leverages learned prototypes to approximate class-conditional feature distributions in the latent space, enabling progressive discrimination through classification sub-tasks of varying granularity. To our knowledge, PDFTime is the first framework to reformulate time series classification as a decoupled, multi-stage similarity-based reasoning process, breaking the long-standing paradigm of direct, black-box feature-to-label mapping. Extensive evaluations demonstrate that PDFTime achieves state-of-the-art (SOTA) performance across UEA and UCR benchmarks. Notably, it secures the top-1 accuracy on 80 out of 128 datasets in the UCR archive, significantly outperforming recent strong baselines in both consistency and generalization.

  • 5 authors
·
May 20

Hyperdimensional Probe: Decoding LLM Representations via Vector Symbolic Architectures

Despite their capabilities, Large Language Models (LLMs) remain opaque with limited understanding of their internal representations. Current interpretability methods, such as direct logit attribution (DLA) and sparse autoencoders (SAEs), provide restricted insight due to limitations such as the model's output vocabulary or unclear feature names. This work introduces Hyperdimensional Probe, a novel paradigm for decoding information from the LLM vector space. It combines ideas from symbolic representations and neural probing to project the model's residual stream into interpretable concepts via Vector Symbolic Architectures (VSAs). This probe combines the strengths of SAEs and conventional probes while overcoming their key limitations. We validate our decoding paradigm with controlled input-completion tasks, probing the model's final state before next-token prediction on inputs spanning syntactic pattern recognition, key-value associations, and abstract inference. We further assess it in a question-answering setting, examining the state of the model both before and after text generation. Our experiments show that our probe reliably extracts meaningful concepts across varied LLMs, embedding sizes, and input domains, also helping identify LLM failures. Our work advances information decoding in LLM vector space, enabling extracting more informative, interpretable, and structured features from neural representations.

  • 5 authors
·
Sep 29, 2025 2

Clinical Concept Centers in LLMs

Large language models are increasingly used in clinical settings. However, research into the reliability and performance of these models has focused almost entirely on the language substrate, scoring what the model says. Mechanistic interpretability has found that the latent space carries a higher fidelity of representation than the text: internal representations not only encode substantially more than the output verbalizes, but the stated reasoning also systematically omits features that causally drive the answer. An evaluation of model behavior in terms of mechanistic interpretability has not been explored in clinical decision support. In this work, we extend behavioral evaluation into the latent space and ask whether clinical concepts exist as locatable, causally used representations inside open-weight LLMs. We find dedicated clinical concept centers in the latent space of all eleven open models we test. These concept centers are interpretable, firing only on their aligned clinical narratives, and meaningfully and causally drive model behavior in both constrained and open-ended settings. They are not just analytical representations, but circuits that can be utilized in clinical practice, and we explore their use from the perspective of both evaluation and performance. From the evaluation standpoint, models stay internally coherent and keep using the relevant concept centers even under adversarial role-based priming, while aligned priming improves downstream clinical performance. From a performance perspective, we simulate realistic deployment settings and find that steering models along these centers leads to meaningful downstream improvements. Finally, we conduct a blinded clinician validation and find the activation and usage of these concept centers predicts clinicians preferences.

  • 5 authors
·
Oct 4

Enhancing Pre-trained Representation Classifiability can Boost its Interpretability

The visual representation of a pre-trained model prioritizes the classifiability on downstream tasks, while the widespread applications for pre-trained visual models have posed new requirements for representation interpretability. However, it remains unclear whether the pre-trained representations can achieve high interpretability and classifiability simultaneously. To answer this question, we quantify the representation interpretability by leveraging its correlation with the ratio of interpretable semantics within the representations. Given the pre-trained representations, only the interpretable semantics can be captured by interpretations, whereas the uninterpretable part leads to information loss. Based on this fact, we propose the Inherent Interpretability Score (IIS) that evaluates the information loss, measures the ratio of interpretable semantics, and quantifies the representation interpretability. In the evaluation of the representation interpretability with different classifiability, we surprisingly discover that the interpretability and classifiability are positively correlated, i.e., representations with higher classifiability provide more interpretable semantics that can be captured in the interpretations. This observation further supports two benefits to the pre-trained representations. First, the classifiability of representations can be further improved by fine-tuning with interpretability maximization. Second, with the classifiability improvement for the representations, we obtain predictions based on their interpretations with less accuracy degradation. The discovered positive correlation and corresponding applications show that practitioners can unify the improvements in interpretability and classifiability for pre-trained vision models. Codes are available at https://github.com/ssfgunner/IIS.

  • 6 authors
·
Oct 28, 2025

SEMASIA: A Large-Scale Dataset of Semantically Structured Latent Representations

Latent representations learned by neural networks often exhibit semantic structure, where concept similarity is reflected by geometric proximity in embedding space. However, comparing such spaces across models remains difficult: changes in architecture, pretraining data, objective, or random seed can yield embeddings with similar content but incompatible geometry. This latent space alignment problem is central to interpretability, transfer and multimodal learning, federated systems, and semantic communication; however, progress remains limited by the lack of large-scale, model-diverse, and metadata-rich benchmarks. To address this gap, we introduce SEMASIA, a large-scale collection of latent representations extracted from approximately 1,700 pretrained vision models across eight standard image-classification benchmarks. SEMASIA pairs embeddings with structured metadata describing architectures, training regimes, pretraining sources, and model scale. We demonstrate three applications of the resource. First, we analyze the conceptual organization of individual latent spaces, showing consistent prototype-like clustering and hierarchical semantic neighborhoods across models and datasets. Second, we benchmark supervised alignment mappings between latent spaces using reconstruction error and downstream task performance. Third, we perform a large-scale regression analysis of how pretraining-data complexity, specialization, transfer learning, augmentation, and model scale relate to geometric and probing properties of embeddings. By coupling representational scale with standardized metadata, SEMASIA provides a reproducible foundation for studying latent geometry, evaluating alignment methods, and developing next-generation heterogeneous and interoperable AI systems.

  • 7 authors
·
May 9

BRIDGE and TCH-Net: Heterogeneous Benchmark and Multi-Branch Baseline for Cross-Domain IoT Botnet Detection

IoT botnet detection has advanced, yet most published systems are validated on a single dataset and rarely generalise across environments. Heterogeneous feature spaces make multi-dataset training practically impossible without discarding semantic interpretability or introducing data integrity violations. No prior work has addressed both problems with a formally specified, reproducible methodology. This paper does. We introduce BRIDGE (Benchmark Reference for IoT Domain Generalisation Evaluation), the first formally specified heterogeneous multi-dataset benchmark for IoT intrusion detection, unifying CICIDS-2017, CIC-IoT-2023, Bot-IoT, Edge-IIoTset, and N-BaIoT through a 46-feature semantic canonical vocabulary grounded in CICFlowMeter nomenclature, with genuine-equivalence-only feature mapping, explicit zero-filling, and per-dataset coverage from 15% to 93%. A leave-one-dataset-out (LODO) protocol makes the generalisation gap precisely measurable: all five evaluated architectures achieve mean LODO F1 between 0.39 and 0.47, and we establish the first community generalisation baseline at mean LODO F1 = 0.5577, a result that shifts the agenda from single-benchmark optimisation toward cross-environment generalisation. We propose TCH-Net, a multi-branch network fusing a three-path Temporal branch (residual convolutional-BiGRU, stride-downsampled BiGRU, pre-LayerNorm Transformer), a provenance-conditioned Contextual branch, and a Statistical branch via Cross-Branch Gated Attention Fusion (CB-GAF) with learnable sigmoid gates for dynamic feature-wise mixing. Across five random seeds, TCH-Net achieves F1 = 0.8296 +/- 0.0028, AUC = 0.9380 +/- 0.0025, and MCC = 0.6972 +/- 0.0056, outperforming all twelve baselines (p < 0.05, Wilcoxon) and recording the highest LODO F1 overall. BRIDGE and the full pipeline are at https://github.com/Ammar-ss/TCH-Net.

  • 7 authors
·
Apr 12

Decomposing MLP Activations into Interpretable Features via Semi-Nonnegative Matrix Factorization

A central goal for mechanistic interpretability has been to identify the right units of analysis in large language models (LLMs) that causally explain their outputs. While early work focused on individual neurons, evidence that neurons often encode multiple concepts has motivated a shift toward analyzing directions in activation space. A key question is how to find directions that capture interpretable features in an unsupervised manner. Current methods rely on dictionary learning with sparse autoencoders (SAEs), commonly trained over residual stream activations to learn directions from scratch. However, SAEs often struggle in causal evaluations and lack intrinsic interpretability, as their learning is not explicitly tied to the computations of the model. Here, we tackle these limitations by directly decomposing MLP activations with semi-nonnegative matrix factorization (SNMF), such that the learned features are (a) sparse linear combinations of co-activated neurons, and (b) mapped to their activating inputs, making them directly interpretable. Experiments on Llama 3.1, Gemma 2 and GPT-2 show that SNMF derived features outperform SAEs and a strong supervised baseline (difference-in-means) on causal steering, while aligning with human-interpretable concepts. Further analysis reveals that specific neuron combinations are reused across semantically-related features, exposing a hierarchical structure in the MLP's activation space. Together, these results position SNMF as a simple and effective tool for identifying interpretable features and dissecting concept representations in LLMs.

  • 3 authors
·
Jun 12, 2025 2

Exploring the encoding of linguistic representations in the Fully-Connected Layer of generative CNNs for Speech

Interpretability work on the convolutional layers of CNNs has primarily focused on computer vision, but some studies also explore correspondences between the latent space and the output in the audio domain. However, it has not been thoroughly examined how acoustic and linguistic information is represented in the fully connected (FC) layer that bridges the latent space and convolutional layers. The current study presents the first exploration of how the FC layer of CNNs for speech synthesis encodes linguistically relevant information. We propose two techniques for exploration of the fully connected layer. In Experiment 1, we use weight matrices as inputs into convolutional layers. In Experiment 2, we manipulate the FC layer to explore how symbolic-like representations are encoded in CNNs. We leverage the fact that the FC layer outputs a feature map and that variable-specific weight matrices are temporally structured to (1) demonstrate how the distribution of learned weights varies between latent variables in systematic ways and (2) demonstrate how manipulating the FC layer while holding constant subsequent model parameters affects the output. We ultimately present an FC manipulation that can output a single segment. Using this technique, we show that lexically specific latent codes in generative CNNs (ciwGAN) have shared lexically invariant sublexical representations in the FC-layer weights, showing that ciwGAN encodes lexical information in a linguistically principled manner.

  • 2 authors
·
Jan 13, 2025

HydraHead: From Head-Level Functional Heterogeneity to Specialized Attention Hybridization

The quadratic complexity of attention poses a critical bottleneck for long-context processing, spurring interest in hybrid attention designs. Most open-source hybrid models adopt a layer-wise strategy. Yet, prior work has noted the inherent difficulty of integrating Linear Attention (LA) with Full Attention (FA), suggesting that the design space of attention hybridization remains underexplored. To probe this space, we conduct interpretability analysis and observe that layers exhibit block-wise functional similarity, while individual heads within the same layer display distinct functional specialization despite sharing input features. This head-level heterogeneity suggests that the head dimension provides a natural and principled granularity for fusing heterogeneous attention signals. Building on this insight, we introduce HydraHead, a novel architecture that hybridizes FA and LA along the head axis. HydraHead features two key innovations: (1) an interpretability-driven selection strategy that identifies retrieval-critical heads and preserves FA only for them, and (2) a scale-normalized fusion module that reconciles the distributional gap between FA and LA head outputs. By leveraging a three-stage transfer pipeline with parameter reuse and distillation, we achieve high-performance hybrid models with minimal training overhead. Under a unified training setup, HydraHead outperforms other hybrid designs in long-context tasks while maintaining strong general reasoning. With interpretability-driven head selection, it matches a 3:1 layer-wise hybrid's long-context performance at a 7:1 LA-to-FA ratio. Crucially, trained on only 15B tokens, HydraHead achieves over 69% improvement over the baseline at 512K context length, approaching Qwen3.5, a leading model of comparable size with a native context length of 256K. This highlights the significant scaling potential of head-level hybridization.

  • 7 authors
·
Jun 17 1

SHaDe: Compact and Consistent Dynamic 3D Reconstruction via Tri-Plane Deformation and Latent Diffusion

We present a novel framework for dynamic 3D scene reconstruction that integrates three key components: an explicit tri-plane deformation field, a view-conditioned canonical radiance field with spherical harmonics (SH) attention, and a temporally-aware latent diffusion prior. Our method encodes 4D scenes using three orthogonal 2D feature planes that evolve over time, enabling efficient and compact spatiotemporal representation. These features are explicitly warped into a canonical space via a deformation offset field, eliminating the need for MLP-based motion modeling. In canonical space, we replace traditional MLP decoders with a structured SH-based rendering head that synthesizes view-dependent color via attention over learned frequency bands improving both interpretability and rendering efficiency. To further enhance fidelity and temporal consistency, we introduce a transformer-guided latent diffusion module that refines the tri-plane and deformation features in a compressed latent space. This generative module denoises scene representations under ambiguous or out-of-distribution (OOD) motion, improving generalization. Our model is trained in two stages: the diffusion module is first pre-trained independently, and then fine-tuned jointly with the full pipeline using a combination of image reconstruction, diffusion denoising, and temporal consistency losses. We demonstrate state-of-the-art results on synthetic benchmarks, surpassing recent methods such as HexPlane and 4D Gaussian Splatting in visual quality, temporal coherence, and robustness to sparse-view dynamic inputs.

  • 1 authors
·
May 22, 2025

Why Does Reinforcement Learning Generalize? A Feature-Level Mechanistic Study of Post-Training in Large Language Models

Reinforcement learning (RL)-based post-training often improves the reasoning performance of large language models (LLMs) beyond the training domain, while supervised fine-tuning (SFT) frequently leads to general capabilities forgetting. However, the mechanisms underlying this contrast remain unclear. To bridge this gap, we present a feature-level mechanistic analysis methodology to probe RL generalization using a controlled experimental setup, where RL- and SFT-tuned models are trained from the same base model on identical data. Leveraging our interpretability framework, we align internal activations across models within a shared feature space and analyze how features evolve during post-training. We find that SFT rapidly introduces many highly specialized features that stabilize early in training, whereas RL induces more restrained and continually evolving feature changes that largely preserve base models' representations. Focusing on samples where RL succeeds but the base model fails, we identify a compact, task-agnostic set of features that directly mediate generalization across diverse tasks. Feature-level interventions confirm their causal role: disabling these features significantly degrades RL models' generalization performance, while amplifying them improves base models' performance. The code is available at https://github.com/danshi777/RL-generalization.

  • 6 authors
·
Apr 26

GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning

Optimization-based latent reasoning improves large language model outputs by optimizing instance-specific continuous states at test time while keeping model parameters frozen. Existing methods, however, typically connect these states to the reasoning trajectory through decoded tokens, making sequence-level credit assignment indirect and obscuring how latent updates shape subsequent reasoning. We introduce GradCuit (gradient through circuit), which inserts optimizable latent states at a selected Transformer layer between the hidden representations of the prompt and the generated continuation. Causal self-attention provides every continuation-token log-probability with a differentiable path to every preceding latent state through the remaining Transformer blocks, enabling reward-weighted gradients from the entire continuation to be assigned directly to the latents. Across five instruction-tuned backbones, three reasoning benchmarks, and two answer formats, GradCuit achieves an average accuracy of 64.5%, outperforming chain-of-thought prompting by 6.6 percentage points and the strongest competing method by 2.4 points. GradCuit also demonstrates greater robustness: across seven learning-rate settings, it consistently outperforms LatentSeek while reducing the standard deviation of accuracy from 1.53 to 0.82, and even its random-walk variant remains competitive with LatentSeek. For interpretability, token-level gradient attribution reveals that latent influence concentrates on reasoning-connector tokens, while layer analysis identifies early-to-middle Transformer layers as the most effective optimization space. By directly optimizing internal reasoning from outcome feedback, GradCuit opens a new axis of robust and interpretable test-time scaling, where LLMs adapt how they reason rather than merely regenerate, sample, or rerank outputs.

  • 7 authors
·
Aug 2 3

R-Capsule: Compressing High-Level Plans for Efficient Large Language Model Reasoning

Chain-of-Thought (CoT) prompting helps Large Language Models (LLMs) tackle complex reasoning by eliciting explicit step-by-step rationales. However, CoT's verbosity increases latency and memory usage and may propagate early errors across long chains. We propose the Reasoning Capsule (R-Capsule), a framework that aims to combine the efficiency of latent reasoning with the transparency of explicit CoT. The core idea is to compress the high-level plan into a small set of learned latent tokens (a Reasoning Capsule) while keeping execution steps lightweight or explicit. This hybrid approach is inspired by the Information Bottleneck (IB) principle, where we encourage the capsule to be approximately minimal yet sufficient for the task. Minimality is encouraged via a low-capacity bottleneck, which helps improve efficiency. Sufficiency is encouraged via a dual objective: a primary task loss for answer accuracy and an auxiliary plan-reconstruction loss that encourages the capsule to faithfully represent the original textual plan. The reconstruction objective helps ground the latent space, thereby improving interpretability and reducing the use of uninformative shortcuts. Our framework strikes a balance between efficiency, accuracy, and interpretability, thereby reducing the visible token footprint of reasoning while maintaining or improving accuracy on complex benchmarks. Our codes are available at: https://anonymous.4open.science/r/Reasoning-Capsule-7BE0

  • 5 authors
·
Sep 26, 2025

MOSAIC: Module Discovery via Sparse Additive Identifiable Causal Learning for Scientific Time Series

Causal representation learning (CRL) seeks to recover latent variables with identifiability guarantees, typically up to permutation and component-wise reparameterization under appropriate assumptions. However, identifiability does not imply interpretability: latent semantics are typically assigned post hoc by alignment with known ground-truth factors. This limitation is particularly acute in scientific time series, where underlying mechanisms are unknown and discovering interpretable structure is a primary goal. In contrast, scientific observations (such as residue-pair distances, climate indices, or process sensors) are inherently semantic, as they correspond to named physical quantities. This raises a key question: can the interpretability of observations be transferred to the identifiable latent space? We propose MOSAIC (Module discovery via Sparse Additive Identifiable Causal learning), a sparse temporal VAE that integrates temporal CRL identifiability with support recovery over observed variables. MOSAIC identifies latent variables via regime-conditioned temporal variation, and recovers for each latent a sparse set of associated observations through an additive decoder, yielding module-level interpretability. We show that ANOVA main-effect supports are identifiable under general smooth mixing functions, and provide finite-sample recovery guarantees for a tractable sparse-additive variant. Empirically, MOSAIC recovers domain-consistent variable groups across RNA molecular dynamics, solar wind, ENSO climate, the Tennessee Eastman process, and a synthetic tokamak benchmark, enabling interpretable discovery of latent mechanisms in scientific time series.

  • 7 authors
·
May 5

From Geometry to Behavior: How Instruction Exposure Unlocks Latent Rhetorical Directions in LLMs

Language models do not merely memorize facts; they absorb rhetorical strategies from pretraining data, including the human tendency to reason by analogy. We ask: is this absorbed capability geometrically accessible before any instruction tuning, and what determines whether it can be activated at inference time? We extract an "Analogy Vector" from the residual stream via difference-in-means and show it enables steerable explanation style across seven instruction-tuned models (Gemma-2, Llama-3.1, Qwen-2.5; 2B-32B), improving harmonic mean from 3.22 to 3.94 (+22%) for Gemma-2-9B-IT on 300 held-out test prompts. A small human validation study confirms the LLM judge on Analogy Quality. Our central finding, from seven matched base-instruct pairs, is a gradient of behavioral accessibility: the analogy direction exists in every base model, but its behavioral accessibility under linear intervention scales continuously with instruction exposure during training. Gemma-2 base models (strict pretraining/IT separation) show flat steering landscapes: present but behaviorally inert. Qwen-2.5 base models (instruction data woven into pretraining) are already partially accessible before any explicit fine-tuning; base-instruct cosine similarity of the analogy vector reaches 0.94-0.97 for Qwen versus 0.37-0.66 for Gemma, providing geometric corroboration of the behavioral gradient. This challenges the binary "base vs instruct" framing: the relevant variable is how much instruction exposure a model has received across its full training pipeline. We position activation steering as a complementary interpretability tool that reveals how instruction exposure organizes rhetorical knowledge in activation space.

  • 1 authors
·
Oct 13

Observable Propagation: A Data-Efficient Approach to Uncover Feature Vectors in Transformers

A key goal of current mechanistic interpretability research in NLP is to find linear features (also called "feature vectors") for transformers: directions in activation space corresponding to concepts that are used by a given model in its computation. Present state-of-the-art methods for finding linear features require large amounts of labelled data -- both laborious to acquire and computationally expensive to utilize. In this work, we introduce a novel method, called "observable propagation" (in short: ObsProp), for finding linear features used by transformer language models in computing a given task -- using almost no data. Our paradigm centers on the concept of observables, linear functionals corresponding to given tasks. We then introduce a mathematical theory for the analysis of feature vectors: we provide theoretical motivation for why LayerNorm nonlinearities do not affect the direction of feature vectors; we also introduce a similarity metric between feature vectors called the coupling coefficient which estimates the degree to which one feature's output correlates with another's. We use ObsProp to perform extensive qualitative investigations into several tasks, including gendered occupational bias, political party prediction, and programming language detection. Our results suggest that ObsProp surpasses traditional approaches for finding feature vectors in the low-data regime, and that ObsProp can be used to better understand the mechanisms responsible for bias in large language models. Code for experiments can be found at github.com/jacobdunefsky/ObservablePropagation.

  • 2 authors
·
Dec 26, 2023

LaDiR: Latent Diffusion Enhances LLMs for Text Reasoning

Large Language Models (LLMs) demonstrate their reasoning ability through chain-of-thought (CoT) generation. However, LLM's autoregressive decoding may limit the ability to revisit and refine earlier tokens in a holistic manner, which can also lead to inefficient exploration for diverse solutions. In this paper, we propose LaDiR (Latent Diffusion Reasoner), a novel reasoning framework that unifies the expressiveness of continuous latent representation with the iterative refinement capabilities of latent diffusion models for an existing LLM. We first construct a structured latent reasoning space using a Variational Autoencoder (VAE) that encodes text reasoning steps into blocks of thought tokens, preserving semantic information and interpretability while offering compact but expressive representations. Subsequently, we utilize a latent diffusion model that learns to denoise a block of latent thought tokens with a blockwise bidirectional attention mask, enabling longer horizon and iterative refinement with adaptive test-time compute. This design allows efficient parallel generation of diverse reasoning trajectories, allowing the model to plan and revise the reasoning process holistically. We conduct evaluations on a suite of mathematical reasoning and planning benchmarks. Empirical results show that LaDiR consistently improves accuracy, diversity, and interpretability over existing autoregressive, diffusion-based, and latent reasoning methods, revealing a new paradigm for text reasoning with latent diffusion.

  • 7 authors
·
Oct 6, 2025

Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistency

Interpreting time series models is uniquely challenging because it requires identifying both the location of time series signals that drive model predictions and their matching to an interpretable temporal pattern. While explainers from other modalities can be applied to time series, their inductive biases do not transfer well to the inherently challenging interpretation of time series. We present TimeX, a time series consistency model for training explainers. TimeX trains an interpretable surrogate to mimic the behavior of a pretrained time series model. It addresses the issue of model faithfulness by introducing model behavior consistency, a novel formulation that preserves relations in the latent space induced by the pretrained model with relations in the latent space induced by TimeX. TimeX provides discrete attribution maps and, unlike existing interpretability methods, it learns a latent space of explanations that can be used in various ways, such as to provide landmarks to visually aggregate similar explanations and easily recognize temporal patterns. We evaluate TimeX on eight synthetic and real-world datasets and compare its performance against state-of-the-art interpretability methods. We also conduct case studies using physiological time series. Quantitative evaluations demonstrate that TimeX achieves the highest or second-highest performance in every metric compared to baselines across all datasets. Through case studies, we show that the novel components of TimeX show potential for training faithful, interpretable models that capture the behavior of pretrained time series models.

  • 6 authors
·
Jun 3, 2023 1

SAE-V: Interpreting Multimodal Models for Enhanced Alignment

With the integration of image modality, the semantic space of multimodal large language models (MLLMs) is more complex than text-only models, making their interpretability more challenging and their alignment less stable, particularly susceptible to low-quality data, which can lead to inconsistencies between modalities, hallucinations, and biased outputs. As a result, developing interpretability methods for MLLMs is crucial for improving alignment quality and efficiency. In text-only LLMs, Sparse Autoencoders (SAEs) have gained attention for their ability to interpret latent representations. However, extending SAEs to multimodal settings presents new challenges due to modality fusion and the difficulty of isolating cross-modal representations. To address these challenges, we introduce SAE-V, a mechanistic interpretability framework that extends the SAE paradigm to MLLMs. By identifying and analyzing interpretable features along with their corresponding data, SAE-V enables fine-grained interpretation of both model behavior and data quality, facilitating a deeper understanding of cross-modal interactions and alignment dynamics. Moreover, by utilizing cross-modal feature weighting, SAE-V provides an intrinsic data filtering mechanism to enhance model alignment without requiring additional models. Specifically, when applied to the alignment process of MLLMs, SAE-V-based data filtering methods could achieve more than 110% performance with less than 50% data. Our results highlight SAE-V's ability to enhance interpretability and alignment in MLLMs, providing insights into their internal mechanisms.

  • 4 authors
·
Feb 22, 2025

Towards mechanistic understanding in a data-driven weather model: internal activations reveal interpretable physical features

Large data-driven physics models like DeepMind's weather model GraphCast have empirically succeeded in parameterizing time operators for complex dynamical systems with an accuracy reaching or in some cases exceeding that of traditional physics-based solvers. Unfortunately, how these data-driven models perform computations is largely unknown and whether their internal representations are interpretable or physically consistent is an open question. Here, we adapt tools from interpretability research in Large Language Models to analyze intermediate computational layers in GraphCast, leveraging sparse autoencoders to discover interpretable features in the neuron space of the model. We uncover distinct features on a wide range of length and time scales that correspond to tropical cyclones, atmospheric rivers, diurnal and seasonal behavior, large-scale precipitation patterns, specific geographical coding, and sea-ice extent, among others. We further demonstrate how the precise abstraction of these features can be probed via interventions on the prediction steps of the model. As a case study, we sparsely modify a feature corresponding to tropical cyclones in GraphCast and observe interpretable and physically consistent modifications to evolving hurricanes. Such methods offer a window into the black-box behavior of data-driven physics models and are a step towards realizing their potential as trustworthy predictors and scientifically valuable tools for discovery.

  • 2 authors
·
Dec 29, 2025

Stage Light is Sequence$^2$: Multi-Light Control via Imitation Learning

Music-inspired Automatic Stage Lighting Control (ASLC) has gained increasing attention in recent years due to the substantial time and financial costs associated with hiring and training professional lighting engineers. However, existing methods suffer from several notable limitations: the low interpretability of rule-based approaches, the restriction to single-primary-light control in music-to-color-space methods, and the limited transferability of music-to-controlling-parameter frameworks. To address these gaps, we propose SeqLight, a hierarchical deep learning framework that maps music to multi-light Hue-Saturation-Value (HSV) space. Our approach first customizes SkipBART, an end-to-end single primary light generation model, to predict the full light color distribution for each frame, followed by hybrid Imitation Learning (IL) techniques to derive an effective decomposition strategy that distributes the global color distribution among individual lights. Notably, the light decomposition module can be trained under varying venue-specific lighting configurations using only mixed light data and no professional demonstrations, thereby flexibly adapting across diverse venues. In this stage, we formulate the light decomposition task as a Goal-Conditioned Markov Decision Process (GCMDP), construct an expert demonstration set inspired by Hindsight Experience Replay (HER), and introduce a three-phase IL training pipeline, achieving strong generalization capability. To validate our IL solution for the proposed GCMDP, we conduct a series of quantitative analysis and human study. The code and trained models are provided at https://github.com/RS2002/SeqLight .

  • 4 authors
·
May 4

Concept-Based Explainable Artificial Intelligence: Metrics and Benchmarks

Concept-based explanation methods, such as concept bottleneck models (CBMs), aim to improve the interpretability of machine learning models by linking their decisions to human-understandable concepts, under the critical assumption that such concepts can be accurately attributed to the network's feature space. However, this foundational assumption has not been rigorously validated, mainly because the field lacks standardised metrics and benchmarks to assess the existence and spatial alignment of such concepts. To address this, we propose three metrics: the concept global importance metric, the concept existence metric, and the concept location metric, including a technique for visualising concept activations, i.e., concept activation mapping. We benchmark post-hoc CBMs to illustrate their capabilities and challenges. Through qualitative and quantitative experiments, we demonstrate that, in many cases, even the most important concepts determined by post-hoc CBMs are not present in input images; moreover, when they are present, their saliency maps fail to align with the expected regions by either activating across an entire object or misidentifying relevant concept-specific regions. We analyse the root causes of these limitations, such as the natural correlation of concepts. Our findings underscore the need for more careful application of concept-based explanation techniques especially in settings where spatial interpretability is critical.

  • 3 authors
·
Jan 31, 2025

Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning

Reasoning models improve their problem-solving ability through inference-time scaling, allocating more compute via longer token budgets. Identifying which reasoning traces are likely to succeed remains a key opportunity: reliably predicting productive paths can substantially reduce wasted computation and improve overall efficiency. We introduce Latent-Trajectory signals that characterize the temporal evolution of a model's internal representations during the generation of intermediate reasoning tokens. By measuring the overall change in latent representations between the start and end of reasoning, the change accumulated across intermediate steps, and the extent to which these changes advance toward the final state, we show that these signals predict solution accuracy more reliably than both cross-layer metrics and output-based confidence measures. When used to guide answer selection across multiple sampled generations, Latent-Trajectory signals make test-time scaling more effective and efficient than majority voting, reducing token usage by up to 70% while preserving and even improving accuracy by 2.6% on average. Moreover, these predictive signals often emerge early in the reasoning trace, enabling early selection and allocation of compute to the most promising candidates. Our findings contribute not only practical strategies for inference-time efficiency, but also a deeper interpretability perspective on how reasoning processes are represented and differentiated in latent space.

MicrosoftResearch Microsoft Research
·
Oct 12, 2025 2

Analysis of Variational Sparse Autoencoders

Sparse Autoencoders (SAEs) have emerged as a promising approach for interpreting neural network representations by learning sparse, human-interpretable features from dense activations. We investigate whether incorporating variational methods into SAE architectures can improve feature organization and interpretability. We introduce the Variational Sparse Autoencoder (vSAE), which replaces deterministic ReLU gating with stochastic sampling from learned Gaussian posteriors and incorporates KL divergence regularization toward a standard normal prior. Our hypothesis is that this probabilistic sampling creates dispersive pressure, causing features to organize more coherently in the latent space while avoiding overlap. We evaluate a TopK vSAE against a standard TopK SAE on Pythia-70M transformer residual stream activations using comprehensive benchmarks including SAE Bench, individual feature interpretability analysis, and global latent space visualization through t-SNE. The vSAE underperforms standard SAE across core evaluation metrics, though excels at feature independence and ablation metrics. The KL divergence term creates excessive regularization pressure that substantially reduces the fraction of living features, leading to observed performance degradation. While vSAE features demonstrate improved robustness, they exhibit many more dead features than baseline. Our findings suggest that naive application of variational methods to SAEs does not improve feature organization or interpretability.

  • 2 authors
·
Sep 26, 2025

Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Process

Despite the growing reasoning capabilities of recent large language models (LLMs), their internal mechanisms during the reasoning process remain underexplored. Prior approaches often rely on human-defined concepts (e.g., overthinking, reflection) at the word level to analyze reasoning in a supervised manner. However, such methods are limited, as it is infeasible to capture the full spectrum of potential reasoning behaviors, many of which are difficult to define in token space. In this work, we propose an unsupervised framework (namely, RISE: Reasoning behavior Interpretability via Sparse auto-Encoder) for discovering reasoning vectors, which we define as directions in the activation space that encode distinct reasoning behaviors. By segmenting chain-of-thought traces into sentence-level 'steps' and training sparse auto-encoders (SAEs) on step-level activations, we uncover disentangled features corresponding to interpretable behaviors such as reflection and backtracking. Visualization and clustering analyses show that these behaviors occupy separable regions in the decoder column space. Moreover, targeted interventions on SAE-derived vectors can controllably amplify or suppress specific reasoning behaviors, altering inference trajectories without retraining. Beyond behavior-specific disentanglement, SAEs capture structural properties such as response length, revealing clusters of long versus short reasoning traces. More interestingly, SAEs enable the discovery of novel behaviors beyond human supervision. We demonstrate the ability to control response confidence by identifying confidence-related vectors in the SAE decoder space. These findings underscore the potential of unsupervised latent discovery for both interpreting and controllably steering reasoning in LLMs.

deepmind Deepmind
·
Dec 30, 2025 3

Learning Geometrically Disentangled Representations of Protein Folding Simulations

Massive molecular simulations of drug-target proteins have been used as a tool to understand disease mechanism and develop therapeutics. This work focuses on learning a generative neural network on a structural ensemble of a drug-target protein, e.g. SARS-CoV-2 Spike protein, obtained from computationally expensive molecular simulations. Model tasks involve characterizing the distinct structural fluctuations of the protein bound to various drug molecules, as well as efficient generation of protein conformations that can serve as an complement of a molecular simulation engine. Specifically, we present a geometric autoencoder framework to learn separate latent space encodings of the intrinsic and extrinsic geometries of the protein structure. For this purpose, the proposed Protein Geometric AutoEncoder (ProGAE) model is trained on the protein contact map and the orientation of the backbone bonds of the protein. Using ProGAE latent embeddings, we reconstruct and generate the conformational ensemble of a protein at or near the experimental resolution, while gaining better interpretability and controllability in term of protein structure generation from the learned latent space. Additionally, ProGAE models are transferable to a different state of the same protein or to a new protein of different size, where only the dense layer decoding from the latent representation needs to be retrained. Results show that our geometric learning-based method enjoys both accuracy and efficiency for generating complex structural variations, charting the path toward scalable and improved approaches for analyzing and enhancing high-cost simulations of drug-target proteins.

  • 5 authors
·
May 20, 2022

LOCOFY Large Design Models -- Design to code conversion solution

Despite rapid advances in Large Language Models and Multimodal Large Language Models (LLMs), numerous challenges related to interpretability, scalability, resource requirements and repeatability remain, related to their application in the design-to-code space. To address this, we introduce the Large Design Models (LDMs) paradigm specifically trained on designs and webpages to enable seamless conversion from design-to-code. We have developed a training and inference pipeline by incorporating data engineering and appropriate model architecture modification. The training pipeline consists of the following: 1)Design Optimiser: developed using a proprietary ground truth dataset and addresses sub-optimal designs; 2)Tagging and feature detection: using pre-trained and fine-tuned models, this enables the accurate detection and classification of UI elements; and 3)Auto Components: extracts repeated UI structures into reusable components to enable creation of modular code, thus reducing redundancy while enhancing code reusability. In this manner, each model addresses distinct but key issues for design-to-code conversion. Separately, our inference pipeline processes real-world designs to produce precise and interpretable instructions for code generation and ensures reliability. Additionally, our models illustrated exceptional end-to-end design-to-code conversion accuracy using a novel preview match score metric. Comparative experiments indicated superior performance of LDMs against LLMs on accuracy of node positioning, responsiveness and reproducibility. Moreover, our custom-trained tagging and feature detection model demonstrated high precision and consistency in identifying UI elements across a wide sample of test designs. Thus, our proposed LDMs are a reliable and superior solution to understanding designs that subsequently enable the generation of efficient and reliable production-ready code.

  • 4 authors
·
Jul 21, 2025