Activation patching is a standard method in mechanistic interpretability for localizing the components of a model responsible for specific behaviors, but it is computationally expensive to apply at scale. Attribution patching offers a faster, gradient-based approximation, yet suffers from noise and reduced reliability in deep, highly non-linear networks. In this work, we introduce Relevance Patching (RelP), which replaces the local gradients in attribution patching with propagation coefficients derived from Layer-wise Relevance Propagation (LRP). LRP propagates the network's output backward through the layers, redistributing relevance to lower-level components according to local propagation rules that ensure properties such as relevance conservation or improved signal-to-noise ratio. Like attribution patching, RelP requires only two forward passes and one backward pass, maintaining computational efficiency while improving faithfulness. We validate RelP across a range of models and tasks, showing that it more accurately approximates activation patching than standard attribution patching, particularly when analyzing residual stream and MLP outputs in the Indirect Object Identification (IOI) task. For instance, for MLP outputs in GPT-2 Large, attribution patching achieves a Pearson correlation of 0.006, whereas RelP reaches 0.956, highlighting the improvement offered by RelP. Additionally, we compare the faithfulness of sparse feature circuits identified by RelP and Integrated Gradients (IG), showing that RelP achieves comparable faithfulness without the extra computational cost associated with IG.

Various XAI attribution methods have been recently proposed for the transformer architecture, allowing for insights into the decision-making process of large language models by assigning importance scores to input tokens and intermediate representations. One class of methods that seems very promising in this direction includes decomposition-based approaches, i.e., XAI-methods that redistribute the model's prediction logit through the network, as this value is directly related to the prediction. In the previous literature we note though that two prominent methods of this category, namely ALTI-Logit and LRP, have not yet been analyzed in juxtaposition and hence we propose to close this gap by conducting a careful quantitative evaluation w.r.t. ground truth annotations on a subject-verb agreement task, as well as various qualitative inspections, using BERT, GPT-2 and LLaMA-3 as a testbed. Along the way we compare and extend the ALTI-Logit and LRP methods, including the recently proposed AttnLRP variant, from an algorithmic and implementation perspective. We further incorporate in our benchmark two widely-used gradient-based attribution techniques. Finally, we make our carefullly constructed benchmark dataset for evaluating attributions on language models, as well as our code, publicly available in order to foster evaluation of XAI-methods on a well-defined common ground.

The field of eXplainable Artificial Intelligence (XAI) aims to bring transparency to today's powerful but opaque deep learning models. While local XAI methods explain individual predictions in form of attribution maps, thereby identifying where important features occur (but not providing information about what they represent), global explanation techniques visualize what concepts a model has generally learned to encode. Both types of methods thus only provide partial insights and leave the burden of interpreting the model's reasoning to the user. In this work we introduce the Concept Relevance Propagation (CRP) approach, which combines the local and global perspectives and thus allows answering both the "where" and "what" questions for individual predictions. We demonstrate the capability of our method in various settings, showcasing that CRP leads to more human interpretable explanations and provides deep insights into the model's representation and reasoning through concept atlases, concept composition analyses, and quantitative investigations of concept subspaces and their role in fine-grained decision making.

Vision transformers (ViTs) can be trained using various learning paradigms, from fully supervised to self-supervised. Diverse training protocols often result in significantly different feature spaces, which are usually compared through alignment analysis. However, current alignment measures quantify this relationship in terms of a single scalar value, obscuring the distinctions between common and unique features in pairs of representations that share the same scalar alignment. We address this limitation by combining alignment analysis with concept discovery, which enables a breakdown of alignment into single concepts encoded in feature space. This fine-grained comparison reveals both universal and unique concepts across different representations, as well as the internal structure of concepts within each of them. Our methodological contributions address two key prerequisites for concept-based alignment: 1) For a description of the representation in terms of concepts that faithfully capture the geometry of the feature space, we define concepts as the most general structure they can possibly form - arbitrary manifolds, allowing hidden features to be described by their proximity to these manifolds. 2) To measure distances between concept proximity scores of two representations, we use a generalized Rand index and partition it for alignment between pairs of concepts. We confirm the superiority of our novel concept definition for alignment analysis over existing linear baselines in a sanity check. The concept-based alignment analysis of representations from four different ViTs reveals that increased supervision correlates with a reduction in the semantic structure of learned representations.

Today's computer vision models achieve human or near-human level performance across a wide variety of vision tasks. However, their architectures, data, and learning algorithms differ in numerous ways from those that give rise to human vision. In this paper, we investigate the factors that affect the alignment between the representations learned by neural networks and human mental representations inferred from behavioral responses. We find that model scale and architecture have essentially no effect on the alignment with human behavioral responses, whereas the training dataset and objective function both have a much larger impact. These findings are consistent across three datasets of human similarity judgments collected using two different tasks. Linear transformations of neural network representations learned from behavioral responses from one dataset substantially improve alignment with human similarity judgments on the other two datasets. In addition, we find that some human concepts such as food and animals are well-represented by neural networks whereas others such as royal or sports-related objects are not. Overall, although models trained on larger, more diverse datasets achieve better alignment with humans than models trained on ImageNet alone, our results indicate that scaling alone is unlikely to be sufficient to train neural networks with conceptual representations that match those used by humans.

The capacity of Large Language Models (LLMs) to follow complex instructions and generate factually accurate text is critical for their real-world application. However, standard decoding methods often fail to robustly satisfy these requirements, while existing control techniques frequently degrade general output quality. In this work, we introduce Attribution-Guided Decoding (AGD), an interpretability-based decoding strategy. Instead of directly manipulating model activations, AGD considers a set of high-probability output token candidates and selects the one that exhibits the highest attribution to a user-defined Region of Interest (ROI). This ROI can be flexibly defined over different parts of the model's input or internal components, allowing AGD to steer generation towards various desirable behaviors. We demonstrate AGD's efficacy across three challenging domains. For instruction following, we show that AGD significantly boosts adherence (e.g., improving the overall success rate on Llama 3.1 from 66.0% to 79.1%). For knowledge-intensive tasks, we show that guiding generation towards usage of internal knowledge components or contextual sources can reduce hallucinations and improve factual accuracy in both closed-book and open-book settings. Furthermore, we propose an adaptive, entropy-based variant of AGD that mitigates quality degradation and reduces computational overhead by applying guidance only when the model is uncertain. Our work presents a versatile, more interpretable, and effective method for enhancing the reliability of modern LLMs.

Multiple instance learning (MIL) is an effective and widely used approach for weakly supervised machine learning. In histopathology, MIL models have achieved remarkable success in tasks like tumor detection, biomarker prediction, and outcome prognostication. However, MIL explanation methods are still lagging behind, as they are limited to small bag sizes or disregard instance interactions. We revisit MIL through the lens of explainable AI (XAI) and introduce xMIL, a refined framework with more general assumptions. We demonstrate how to obtain improved MIL explanations using layer-wise relevance propagation (LRP) and conduct extensive evaluation experiments on three toy settings and four real-world histopathology datasets. Our approach consistently outperforms previous explanation attempts with particularly improved faithfulness scores on challenging biomarker prediction tasks. Finally, we showcase how xMIL explanations enable pathologists to extract insights from MIL models, representing a significant advance for knowledge discovery and model debugging in digital histopathology. Codes are available at: this https URL

The field of mechanistic interpretability aims to study the role of individual neurons in Deep Neural Networks. Single neurons, however, have the capability to act polysemantically and encode for multiple (unrelated) features, which renders their interpretation difficult. We present a method for disentangling polysemanticity of any Deep Neural Network by decomposing a polysemantic neuron into multiple monosemantic "virtual" neurons. This is achieved by identifying the relevant sub-graph ("circuit") for each "pure" feature. We demonstrate how our approach allows us to find and disentangle various polysemantic units of ResNet models trained on ImageNet. While evaluating feature visualizations using CLIP, our method effectively disentangles representations, improving upon methods based on neuron activations. Our code is available at this https URL.

Artificial intelligence has started to transform histopathology impacting clinical diagnostics and biomedical research. However, while many computational pathology approaches have been proposed, most current AI models are limited with respect to generalization, application variety, and handling rare diseases. Recent efforts introduced self-supervised foundation models to address these challenges, yet existing approaches do not leverage pathologist knowledge by design. In this study, we present a novel approach to designing foundation models for computational pathology, incorporating pathologist expertise, semi-automated data curation, and a diverse dataset from over 15 laboratories, including 58 tissue types, and encompassing 129 different histochemical and immunohistochemical staining modalities. We demonstrate that our model "RudolfV" surpasses existing state-of-the-art foundation models across different benchmarks focused on tumor microenvironment profiling, biomarker evaluation, and reference case search while exhibiting favorable robustness properties. Our study shows how domain-specific knowledge can increase the efficiency and performance of pathology foundation models and enable novel application areas.

Diffusion models have demonstrated remarkable capabilities in synthesizing realistic images, spurring interest in using their representations for various downstream tasks. To better understand the robustness of these representations, we analyze popular Stable Diffusion models using representational similarity and norms. Our findings reveal three phenomena: (1) the presence of a learned positional embedding in intermediate representations, (2) high-similarity corner artifacts, and (3) anomalous high-norm artifacts. These findings underscore the need to further investigate the properties of diffusion model representations before considering them for downstream tasks that require robust features. Project page: this https URL

Gradient-based optimization has been a cornerstone of machine learning that enabled the vast advances of Artificial Intelligence (AI) development over the past decades. However, this type of optimization requires differentiation, and with recent evidence of the benefits of non-differentiable (e.g. neuromorphic) architectures over classical models w.r.t. efficiency, such constraints can become limiting in the future. We present Layer-wise Feedback Propagation (LFP), a novel training principle for neural network-like predictors that utilizes methods from the domain of explainability to decompose a reward to individual neurons based on their respective contributions. Leveraging these neuron-wise rewards, our method then implements a greedy approach reinforcing helpful parts of the network and weakening harmful ones. While having comparable computational complexity to gradient descent, LFP does not require gradient computation and generates sparse and thereby memory- and energy-efficient parameter updates and models. We establish the convergence of LFP theoretically and empirically, demonstrating its effectiveness on various models and datasets. Via two applications - neural network pruning and the approximation-free training of Spiking Neural Networks (SNNs) - we demonstrate that LFP combines increased efficiency in terms of computation and representation with flexibility w.r.t. choice of model architecture and objective function. Our code is available at this https URL.

From implicit differentiation to probabilistic modeling, Jacobian and Hessian matrices have many potential use cases in Machine Learning (ML), but they are viewed as computationally prohibitive. Fortunately, these matrices often exhibit sparsity, which can be leveraged to speed up the process of Automatic Differentiation (AD). This paper presents advances in sparsity detection, previously the performance bottleneck of Automatic Sparse Differentiation (ASD). Our implementation of sparsity detection is based on operator overloading, able to detect both local and global sparsity patterns, and supports flexible index set representations. It is fully automatic and requires no modification of user code, making it compatible with existing ML codebases. Most importantly, it is highly performant, unlocking Jacobians and Hessians at scales where they were considered too expensive to compute. On real-world problems from scientific ML, graph neural networks and optimization, we show significant speed-ups of up to three orders of magnitude. Notably, using our sparsity detection system, ASD outperforms standard AD for one-off computations, without amortization of either sparsity detection or matrix coloring.