Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performance

Deep neural networks (DNNs) are widely used, but interpreting what they actually learn remains difficult. A major obstacle is that individual neurons often encode multiple unrelated concepts, obscuring the decision process of the network. While prior work, such as sparse autoencoders, can separate these mixed signals into more meaningful, "monosemantic" features, this typically requires altering the model in ways that can degrade downstream performance. To overcome this, we introduce ELUDe (explicit, lossless, unsupervised disentanglement), a method for improving the interpretability of DNNs while preserving their functional equivalence. ELUDe breaks latent representations into clear, inspectable sub-units that behave like interpretable features, while guaranteeing that the model's outputs remain exactly the same. It requires no explicit training, no labels, and can be applied to pretrained models. ELUDe works by reorganizing how information flows between layers, re-routing concept-specific contributions while preserving the original computation by construction. Across several vision models, including DINOv2 and supervised ViT-B/16, ELUDe improves interpretability, keeps downstream accuracy unchanged, runs efficiently, and supports practical uses such as steering model representations. In short, ELUDe offers interpretability (almost) without a tradeoff: clearer, scalable, and actionable model insights with no loss in performance.

Seeing Through Circuits: Faithful Mechanistic Interpretability for Vision Transformers

Transparency of neural networks' internal reasoning is at the heart of interpretability research, adding to trust, safety, and understanding of these models. The field of mechanistic interpretability has recently focused on studying task-specific computational graphs, defined by connections (edges) between model components. Such edge-based circuits have been defined in the context of large language models, yet vision-based approaches so far only consider neuron-based circuits. These tell which information is encoded, but not how it is routed through the complex wiring of a neural network. In this work, we investigate whether useful mechanistic circuits can be identified through computational graphs in vision transformers. We propose an effective method for Automatic Visual Circuit Discovery (Vi-CD) that recovers class-specific circuits for classification, identifies circuits underlying typographic attacks in CLIP, and discovers circuits that lend themselves for steering to correct harmful model behavior. Overall, we find that insightful and actionable edge-based circuits can be recovered from vision transformers, adding transparency to the internal computations of these models.

Certified Circuits: Stability Guarantees for Mechanistic Circuits

Understanding how neural networks arrive at their predictions is essential for debugging, auditing, and deployment. Mechanistic interpretability pursues this goal by identifying circuits - minimal subnetworks responsible for specific behaviors. However, existing circuit discovery methods are brittle: circuits depend strongly on the chosen concept dataset and often fail to transfer out-of-distribution, raising doubts whether they capture concept or dataset-specific artifacts. We introduce Certified Circuits, which provide provable stability guarantees for circuit discovery. Our framework wraps any black-box discovery algorithm with randomized data subsampling to certify that circuit component inclusion decisions are invariant to bounded edit-distance perturbations of the concept dataset. Unstable neurons are abstained from, yielding circuits that are more compact and more accurate. On ImageNet and OOD datasets, certified circuits achieve up to 91% higher accuracy while using 45% fewer neurons, and remain reliable where baselines degrade. Certified Circuits puts circuit discovery on formal ground by producing mechanistic explanations that are provably stable and better aligned with the target concept. Code will be released soon!

Insight: Interpretable Semantic Hierarchies in Vision-Language Encoders

Language-aligned vision foundation models perform strongly across diverse downstream tasks. Yet, their learned representations remain opaque, making interpreting their decision-making hard. Recent works decompose these representations into human-interpretable concepts, but provide poor spatial grounding and are limited to image classification tasks. In this work, we propose Insight, a language-aligned concept foundation model that provides fine-grained concepts, which are human-interpretable and spatially grounded in the input image. We leverage a hierarchical sparse autoencoder and a foundation model with strong semantic representations to automatically extract concepts at various granularities. Examining local co-occurrence dependencies of concepts allows us to define concept relationships. Through these relations we further improve concept naming and obtain richer explanations. On benchmark data, we show that Insight provides performance on classification and segmentation that is competitive with opaque foundation models while providing fine-grained, high quality concept-based explanations. Code is available at https://github.com/kawi19/Insight.

Temporal Concept Dynamics in Diffusion Models via Prompt-Conditioned Interventions

Diffusion models are usually evaluated by their final outputs, gradually denoising random noise into meaningful images. Yet, generation unfolds along a trajectory, and analyzing this dynamic process is crucial for understanding how controllable, reliable, and predictable these models are in terms of their success/failure modes. In this work, we ask the question: when does noise turn into a specific concept (e.g., age) and lock in the denoising trajectory? We propose PCI (Prompt-Conditioned Intervention) to study this question. PCI is a training-free and model-agnostic framework for analyzing concept dynamics through diffusion time. The central idea is the analysis of Concept Insertion Success (CIS), defined as the probability that a concept inserted at a given timestep is preserved and reflected in the final image, offering a way to characterize the temporal dynamics of concept formation. Applied to several state-of-the-art text-to-image diffusion models and a broad taxonomy of concepts, PCI reveals diverse temporal behaviors across diffusion models, in which certain phases of the trajectory are more favorable to specific concepts even within the same concept type. These findings also provide actionable insights for text-driven image editing, highlighting when interventions are most effective without requiring access to model internals or training, and yielding quantitatively stronger edits that achieve a balance of semantic accuracy and content preservation than strong baselines. Code is available at: https://github.com/adagorgun/PCI-Prompt-Controlled-Interventions

Disentangling Polysemantic Channels in Convolutional Neural Networks

Mechanistic interpretability is concerned with analyzing individual components in a (convolutional) neural network (CNN) and how they form larger circuits representing decision mechanisms. These investigations are challenging since CNNs frequently learn polysemantic channels that encode distinct concepts, making them hard to interpret. To address this, we propose an algorithm to disentangle a specific kind of polysemantic channel into multiple channels, each responding to a single concept. Our approach restructures weights in a CNN, utilizing that different concepts within the same channel exhibit distinct activation patterns in the previous layer. By disentangling these polysemantic features, we enhance the interpretability of CNNs, ultimately improving explanatory techniques such as feature visualizations.

VITAL: More Understandable Feature Visualization through Distribution Alignment and Relevant Information Flow

Neural networks are widely adopted to solve complex and challenging tasks. Especially in high-stakes decision-making, understanding their reasoning process is crucial, yet proves challenging for modern deep networks. Feature visualization (FV) is a powerful tool to decode what information neurons are responding to and hence to better understand the reasoning behind such networks. In particular, in FV we generate human-understandable images that reflect the information detected by neurons of interest. However, current methods often yield unrecognizable visualizations, exhibiting repetitive patterns and visual artifacts that are hard to understand for a human. To address these problems, we propose to guide FV through statistics of real image features combined with measures of relevant network flow to generate prototypical images. Our approach yields human-understandable visualizations that both qualitatively and quantitatively improve over state-of-the-art FVs across various architectures. As such, it can be used to decode which information the network uses, complementing mechanistic circuits that identify where it is encoded. Code is available at: https://github.com/adagorgun/VITAL

Escaping Plato's Cave: Robust Conceptual Reasoning through Interpretable 3D Neural Object Volumes

With the rise of neural networks, especially in high-stakes applications, these networks need two properties (i) robustness and (ii) interpretability to ensure their safety. Recent advances in classifiers with 3D volumetric object representations have demonstrated a greatly enhanced robustness in out-of-distribution data. However, these 3D-aware classifiers have not been studied from the perspective of interpretability. We introduce CAVE - Concept Aware Volumes for Explanations - a new direction that unifies interpretability and robustness in image classification. We design an inherently-interpretable and robust classifier by extending existing 3D-aware classifiers with concepts extracted from their volumetric representations for classification. In an array of quantitative metrics for interpretability, we compare against different concept-based approaches across the explainable AI literature and show that CAVE discovers well-grounded concepts that are used consistently across images, while achieving superior robustness.

Unlocking Open-Set Language Accessibility in Vision Models

Visual classifiers offer high-dimensional feature representations that are challenging to interpret and analyze. Text, in contrast, provides a more expressive and human-friendly interpretable medium for understanding and analyzing model behavior. We propose a simple, yet powerful method for reformulating any visual classifier so that it can be accessed with open-set text queries without compromising its original performance. Our approach is label-free, efficient, and preserves the underlying classifier's distribution and reasoning processes. We thus unlock several text-based interpretability applications for any classifier. We apply our method on 40 visual classifiers and demonstrate two primary applications: 1) building both label-free and zero-shot concept bottleneck models and therefore converting any classifier to be inherently-interpretable and 2) zero-shot decoding of visual features into natural language. In both applications, we achieve state-of-the-art results, greatly outperforming existing works. Our method enables text approaches for interpreting visual classifiers.

Now you see me! A framework for obtaining class-relevant saliency maps

Neural networks are part of daily-life decision-making, including in high-stakes settings where understanding and transparency are key. Saliency maps have been developed to gain understanding into which input features neural networks use for a specific prediction. Although widely employed, these methods often result in overly general saliency maps that fail to identify the specific information that triggered the classification. In this work, we suggest a framework that allows to incorporate attributions across classes to arrive at saliency maps that actually capture the class-relevant information. On established benchmarks for attribution methods, including the grid-pointing game and randomization-based sanity checks, we show that our framework heavily boosts the performance of standard saliency map approaches. It is, by design, agnostic to model architectures and attribution methods and now allows to identify the distinguishing and shared features used for a model prediction.