On the Properties of Feature Attribution for Supervised Contrastive Learning

While Contrastive Learning (CL) has revolutionized self-supervised representation learning, its latent representations remain highly entangled and opaque, limiting their interpretability in safety-critical applications. We identify that a fundamental cause of this entanglement is the reliance on deterministic similarity measures, which treat all feature dimensions equally. In compositional scenes, this creates an Optimization Conflict: common background features, such as, "blue sky", are encouraged to align in positive pairs but simultaneously repelled in negative pairs, causing gradient oscillations that hinder precise semantic disentanglement. To address this, we propose BayesNCL (Bayesian Gated Non-Negative Contrastive Learning). Unlike standard approaches, BayesNCL introduces a probabilistic gating mechanism that dynamically filters out task-irrelevant, high-frequency common features while selectively retaining discriminative semantics. By formalizing feature selection as a variational inference problem with a sparse Bernoulli prior, our method effectively resolves the optimization conflict. Empirical experimental results on Imagenet-100 demonstrate that BayesNCL achieves a remarkable 142.1% improvement in semantic consistency compared to state-of-the-art baselines, yielding highly interpretable representations without compromising downstream task performance. Code is available at https://github.com/Cui-Peng-624/BayesNCL.

Neuron labeling assigns textual descriptions to internal units of deep networks. Existing approaches typically rely on highly activating examples, often yielding broad or misleading labels by focusing on dominant but incidental visual factors. Prior work such as FALCON introduced contrastive examples -- inputs that are semantically similar to activating examples but elicit low activations -- to sharpen explanations, but it primarily addresses subspace-level interpretability rather than scalable neuron-level labeling. We revisit contrastive explanations for neuron-level labeling in two stages: (1) candidate label generation with vision language models (VLMs) and (2) label assignment with CLIP-like encoders. First, we show that providing contrastive image sets to VLMs yields candidate labels that are more specific and more faithful. Second, we introduce Contrastive Semantic Projection (CSP), an extension of SemanticLens that incorporates contrastive examples directly into its CLIP-based scoring and selection pipeline. Across extensive experiments and a case study on melanoma detection, contrastive labeling improves both faithfulness and semantic granularity over state-of-the-art baselines. Our results demonstrate that contrastive examples are a simple yet powerful and currently underutilized component of neuron labeling and analysis pipelines.

Occlusion-based attribution methods provide an intuitive way to estimate feature importance by perturbing input features and measuring the resulting change in model output. However, their reliability is strongly affected by how feature removal is implemented: externally selected baselines can introduce bias, out-of-distribution samples, and unstable explanations, while in nonlinear models the occlusion of a set of features can also alter the contribution of non-occluded features. We refer to this effect as attribution shift, as the attribution scores of the non-occluded features drift from their initial values. To challenge these major issues that render explanations unstable, we introduce XtrAIn, a training-guided attribution method that transfers the occlusion operation from the input space to the parameter space. Instead of replacing input values with hand-crafted baselines, XtrAIn follows the model's training trajectory and measures how feature-associated parameter updates affect the output logits. We further introduce Xstep, a lightweight approximation for reducing computational cost, and XtrAIn+, a target-focused variant that emphasizes updates aligned with the target class. Experiments on controlled image datasets and PAM50 breast-cancer subtype classification show that the proposed methods produce cleaner and more interpretable attribution patterns than standard attribution baselines. Overall, XtrAIn provides a training-aware perspective on feature attribution and offers a useful diagnostic tool for studying how feature-level evidence is formed during training.