This paper studies domain generalization in an anti-causal setting where the outcome causes the covariates. The authors propose two methods that leverage unlabeled data from multiple environments to regularize the model's sensitivity to changes in the mean and covariance of covariates, with worst-case optimality guarantees. Empirical results are shown on a controlled physical system and a physiological signal dataset.

Reasoning Large Language Models (LLMs) benefit from test-time scaling, with accuracy improving as token budget increases, motivating adaptive reasoning. However, setting token budgets and thresholds involves a risk-accuracy trade-off. This paper re-frames budget setting as risk control, limiting error rate while minimizing compute. It introduces an upper threshold to stop when confident and a lower threshold to stop unsolvable instances. Using distribution-free risk control, the framework optimally specifies stopping mechanisms given a target risk and validation set. Experiments across diverse tasks and models demonstrate computational efficiency gains while adhering to risk targets.

Residual Context Diffusion (RCD) is a new module for diffusion large language models (dLLMs) that recycles computation from discarded tokens, improving efficiency and accuracy. RCD converts discarded token representations into contextual residuals and reinjects them into the denoising process. It uses a decoupled two-stage training pipeline and achieves 5-10 point accuracy gains across benchmarks, with up to 4-5x fewer denoising steps on challenging tasks like AIME.

Diffusion Large Language Models (dLLMs) now match autoregressive models on many tasks while being more efficient. A key design aspect is the sampling procedure that selects which tokens to unmask. Current heuristics require manual tuning and degrade with larger block sizes. This work proposes training sampling policies via reinforcement learning, formalizing masked diffusion sampling as a Markov decision process, and using a lightweight single-layer transformer policy. Experiments show the trained policies match state-of-the-art heuristics in semi-autoregressive (block) generation and outperform them in full-diffusion settings.

MemoryLLM decouples feed-forward modules from self-attention, enabling interpretable context-free token-wise neural retrieval memory and improving inference efficiency.

Reinforcement learning finetuning improves vision-language models on reasoning benchmarks, but they remain vulnerable to textual perturbations. This paper shows that misleading captions or incorrect chain-of-thought traces cause significant drops in robustness, especially when consistency is considered. Closed models outperform open-source ones, indicating a gap in current open-source RL finetuning. An accuracy-faithfulness trade-off is identified; adversarial augmentation alone is insufficient, and a faithfulness-aware reward can help but may lead to shortcut learning.

This paper proposes amortized MIPS, a regression-based approach that trains neural networks to directly predict MIPS solutions. The key insight is that the MIPS value function is the support function of the key set, whose gradient yields the optimal key. Two models are introduced: SupportNet, an input-convex network for regressing the support function, and KeyNet, a vector-valued network for directly regressing the optimal key. Experiments on BEIR show improved IVF hit rates under equal compute budgets.

Self-organizing LLM teams fail to match the performance of their best expert agent, with losses up to 41.1%. The primary bottleneck is leveraging expert knowledge rather than identifying experts. Teams tend toward integrative compromise, which harms performance but improves robustness against adversarial agents.

Distributed GPU training at scale encounters three failure modes: crashed jobs, silent slowdowns, and numerical corruption. Databricks AI uses stress testing with diverse workloads and a multi-stage health check system (active bootstrap, passive continuous, periodic multi-node) to catch failures early and maintain reliability.