University of Notre Dame

Reinforcement learning with verifiable rewards (RLVR) trains a language model on problems that may themselves be confidential, and the released weights can reveal which problems it saw. DP-GRPO aggregates all responses to one prompt into a single gradient, clips it once and adds Gaussian noise, so the privacy budget depends on neither the number of responses per prompt nor the clipping norm; the authors describe it as the first differential privacy guarantee for RLVR training. Training Qwen2.5-1.5B-Instruct with LoRA at a per-run budget of 8, the reward signal beats a control that removes only the reward by 2.65 points on MATH and 3.24 on GSM8K, positive in every seed.

Reinforcement learning with verifiable rewards (RLVR) trains a language model on problems that may themselves be confidential, and the released weights can reveal which problems it saw. DP-GRPO aggregates all responses to one prompt into a single gradient, clips it once and adds Gaussian noise, so the privacy budget depends on neither the number of responses per prompt nor the clipping norm; the authors describe it as the first differential privacy guarantee for RLVR training. Training Qwen2.5-1.5B-Instruct with LoRA at a per-run budget of 8, the reward signal beats a control that removes only the reward by 2.65 points on MATH and 3.24 on GSM8K, positive in every seed.

With only black-box access to a recommendation API, injecting under 1% synthetic interactions makes a generative recommender promote an attacker's target far more often whenever a chosen trigger item appears, and short prefixes picked for decoding alignment raise the hit-rate lift by up to 16.3 percentage points.

With only black-box access to a recommendation API, injecting under 1% synthetic interactions makes a generative recommender promote an attacker's target far more often whenever a chosen trigger item appears, and short prefixes picked for decoding alignment raise the hit-rate lift by up to 16.3 percentage points.

Across 18 Base–Instruct model pairs, harmfulness is more decodable from attention updates while refusal evidence is written mainly by MLPs, and post-training strengthens that writing without improving recognition.

Across 18 Base–Instruct model pairs, harmfulness is more decodable from attention updates while refusal evidence is written mainly by MLPs, and post-training strengthens that writing without improving recognition.

The work introduces BHyGNN+, a self-supervised framework for heterophilic hypergraph representation learning that contrasts augmented views of a hypergraph against its dual (with node and hyperedge roles interchanged) using cosine similarity, learning representations without ground-truth labels and without negative samples, and achieving node-classification accuracy above supervised and self-supervised baselines on eleven benchmark datasets covering heterophilic and homophilic hypergraphs plus a synthetic heterophilic dataset.

The work introduces BHyGNN+, a self-supervised framework for heterophilic hypergraph representation learning that contrasts augmented views of a hypergraph against its dual (with node and hyperedge roles interchanged) using cosine similarity, learning representations without ground-truth labels and without negative samples, and achieving node-classification accuracy above supervised and self-supervised baselines on eleven benchmark datasets covering heterophilic and homophilic hypergraphs plus a synthetic heterophilic dataset.

This work introduces FiberPro 1.0, a large language model-driven multi-agent framework that couples high-throughput focused rotary jet spinning (FRJS) with real-time experimental feedback, converging on spinnable formulations within an average of about two iterations for three proteins with no prior FRJS precedent (soy protein isolate, keratin, bovine serum albumin), and uses it to design a zein-poly(ethylene oxide)-curcumin (Zein-PC) active coating that deposits conformally on diverse produce and combines antibacterial and antioxidant activity, pH-responsive colorimetric sensing, and image-based shelf-life prediction.

This work introduces FiberPro 1.0, a large language model-driven multi-agent framework that couples high-throughput focused rotary jet spinning (FRJS) with real-time experimental feedback, converging on spinnable formulations within an average of about two iterations for three proteins with no prior FRJS precedent (soy protein isolate, keratin, bovine serum albumin), and uses it to design a zein-poly(ethylene oxide)-curcumin (Zein-PC) active coating that deposits conformally on diverse produce and combines antibacterial and antioxidant activity, pH-responsive colorimetric sensing, and image-based shelf-life prediction.