• Treat diffusion denoising as state estimation, not ODE solving. DiFA borrows the Kalman filter idea, fusing past predictions along the reverse trajectory to correct each step. It's training-free, drops onto existing models, and improves several metrics on CIFAR-10 and ImageNet — though the abstract only offers qualitative claims.

• Stop collapsing the reviewer's feedback into a single score. LLM-as-a-Coach distills a judge's rubric verdicts into transferable "experiential knowledge" and feeds it back to the policy, aimed squarely at writing and dialogue tasks where correctness can't be verified and GRPO/RLHF struggle most.

• One wearable camera captures both the viewer and the view. ReViV reconstructs a 4D representation of the viewer (camera/gaze/body/hands) and the view (scene depth) in a single feed-forward pass from monocular egocentric video, hits SOTA on several benchmarks, and ships code and models fully open.

Also Notable

• Princeton's RL Survey, From Classic Algorithms to the Foundation-Model Era — good for anyone who wants a systematic pass over RL and how it became today's post-training workhorse.
• Using CoT Reasoning to Stabilize Value Conflicts in RLHF — a geometric view of why scalar rewards leave models wavering between conflicting values.
• Mamba/State-Space Models as Adaptive Neural Operators for PDEs — solving on arbitrary geometries and varied meshes (ICLR).
• Industrial Video Anomaly Detection With Object-Centric Tracking Plus VLM Reasoning — targets how general VLMs miss detections and can't explain anomalies in real industrial settings.
• Generating Routable Pedestrian Networks From Aerial Imagery — the hard part isn't spotting sidewalks but completing occluded, implicitly defined connectivity, handled with RL.

Read the full edition →