Do Clinical Models Change Treatment Decisions?

Clinical foundation models are evaluated with factual or exam-style medical QA, but treatment decisions must change when patient context changes. We introduce ClinPivot, an auditable treatment-decision benchmark built from biomedical relations and pivoted patient contexts. ClinPivot asks whether models change treatment choices when new clinical constraints shift the action space. We find that strong medical QA performance does not reliably predict decision-making performance: frontier models and task-adapted Qwen variants often fail to change decisions correctly, and model rankings shift across evaluation regimes. Decision-structured supervision improves pivot-sensitive decision-making and medical QA under matched knowledge budgets, while lightweight replay reduces losses in general assistant ability.

Forgetting in Language Models: Capacity, Optimization, and Self-Generated Replay

Models trained on a new task typically degrade on prior tasks, a phenomenon known as forgetting. Traditionally, mitigating forgetting has required replaying stored exemplars from prior tasks, which is often impractical. By contrast, language models can sample from their own training distribution, and we show that these self-generated samples serve as effective replay data, nearly eliminating forgetting. We find that forgetting nonetheless persists when the model has little remaining capacity: models pretrained close to saturation cannot absorb new information without overwriting prior knowledge. When capacity is not the limiting factor, low learning rates reduce forgetting but require substantially more training steps. Replay breaks this tradeoff, enabling fast, high-learning-rate finetuning without forgetting.

Dealing with the Evil Twins: Improving Random Augmentation by Addressing Catastrophic Forgetting of Diverse Augmentations

Data augmentation is a promising tool for enhancing out-of-distribution generalization, where the key is to produce diverse, challenging variations of the source domain via costly targeted augmentations that maximize its generalization effect. Conversely, random augmentation is inexpensive but is deemed suboptimal due to its limited effect. In this paper, we revisit random augmentation and explore methods to address its shortcomings. We show that the stochastic nature of random augmentation can produce a set of colliding augmentations that distorts the learned features, similar to catastrophic forgetting. We propose a simple solution that improves the generalization effect of random augmentation by addressing forgetting, which displays strong generalization performance across various single source domain generalization (sDG) benchmarks.

Cost-Efficient Continual Learning with Sufficient Exemplar Memory

Continual learning (CL) research typically assumes highly constrained exemplar memory resources. However, in many real-world scenarios-especially in the era of large foundation models-memory is abundant, while GPU computational costs are the primary bottleneck. In this work, we investigate CL in a novel setting where exemplar memory is ample (i.e., sufficient exemplar memory). Unlike prior methods designed for strict exemplar memory constraints, we propose a simple yet effective approach that directly operates in the model's weight space through a combination of weight resetting and averaging techniques. Our method achieves state-of-the-art performance while reducing the computational cost to a quarter or third of existing methods. These findings challenge conventional CL assumptions and provide a practical baseline for computationally efficient CL applications.