Search

Unlike human-engineered systems such as aeroplanes, where each component's role and dependencies are well understood, the inner workings of AI models remain largely opaque, hindering verifiability and undermining trust. This paper introduces SemanticLens, a universal explanation method for neural networks that maps hidden knowledge encoded by components (e.g., individual neurons) into the semantically structured, multimodal space of a foundation model such as CLIP. In this space, unique operations become possible, including (i) textual search to identify neurons encoding specific concepts, (ii) systematic analysis and comparison of model representations, (iii) automated labelling of neurons and explanation of their functional roles, and (iv) audits to validate decision-making against requirements. Fully scalable and operating without human input, SemanticLens is shown to be effective for debugging and validation, summarizing model knowledge, aligning reasoning with expectations (e.g., adherence to the ABCDE-rule in melanoma classification), and detecting components tied to spurious correlations and their associated training data. By enabling component-level understanding and validation, the proposed approach helps bridge the "trust gap" between AI models and traditional engineered systems. We provide code for SemanticLens on https://github.com/jim-berend/semanticlens and a demo on https://semanticlens.hhi-research-insights.eu., Comment: 74 pages (18 pages manuscript, 7 pages references, 49 pages appendix)

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., Comment: 19 pages, 17 figures, code: https://github.com/jvielhaben/NLMCD-ALIGN

Gradient-based optimization has been a cornerstone of machine learning enabling the vast advances of AI development over the past decades. However, since this type of optimization requires differentiation, it reduces flexibility in the choice of model and objective. With recent evidence of the benefits of non-differentiable (e.g. neuromorphic) architectures over classical models, such constraints can become limiting in the future. We present Layer-wise Feedback Propagation (LFP), a novel training principle for neural network-like predictors utilizing methods from the domain of explainability to decompose a reward to individual neurons based on their respective contributions to solving a given task without imposing any differentiability requirements. 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 offers the advantage that it obtains sparse models due to an implicit weight scaling. We establish the convergence of LFP theoretically and empirically, demonstrating its effectiveness on various models and datasets. We further investigate two applications for LFP: Firstly, neural network pruning, and secondly, the optimization of neuromorphic architectures such as Heaviside step function activated Spiking Neural Networks (SNNs). In the first setting, LFP naturally generates sparse models that are easily prunable and thus efficiently encode and compute information. In the second setting, LFP achieves comparable performance to surrogate gradient descent, but provides approximation-free training, which eases the implementation on neuromorphic hardware. Consequently, LFP combines efficiency in terms of computation and representation with flexibility w.r.t. model architecture and objective function. Our code is available.