University of Georgia

In streaming deployment a fixed policy's loss can drift with observed history, so certifying a fixed population risk no longer describes the risk actually encountered. StreamLTT certifies average predictable risk, the conditional risk of each decision given the information available before its outcome, averaged along the stream, and conditional centering turns empirical deviations into a martingale sum. Across seven forecasting datasets and seven deployment risks, its variance-adaptive version pushes the certified frontier close to the risk target while empirical family-wise error stays near the prescribed level.

In streaming deployment a fixed policy's loss can drift with observed history, so certifying a fixed population risk no longer describes the risk actually encountered. StreamLTT certifies average predictable risk, the conditional risk of each decision given the information available before its outcome, averaged along the stream, and conditional centering turns empirical deviations into a martingale sum. Across seven forecasting datasets and seven deployment risks, its variance-adaptive version pushes the certified frontier close to the risk target while empirical family-wise error stays near the prescribed level.

Structured knowledge retrieval no longer needs predefined query-rule pairs: a routing agent picks a reasoning rule per query, its retriever ranks evidence under that rule's constraints, and the two train in alternation, each supervising the other.

Structured knowledge retrieval no longer needs predefined query-rule pairs: a routing agent picks a reasoning rule per query, its retriever ranks evidence under that rule's constraints, and the two train in alternation, each supervising the other.

TopoLIA extends attention from how much information is exchanged to how that information is transformed per target, beating its standard-attention counterpart on protein–ligand scoring and docking benchmarks and reaching an 88% docking success rate on CASF-2013.

TopoLIA extends attention from how much information is exchanged to how that information is transformed per target, beating its standard-attention counterpart on protein–ligand scoring and docking benchmarks and reaching an 88% docking success rate on CASF-2013.

MolP2SBench recasts molecular property-to-structure inference as masked-scaffold multiple choice and adds shuffled and counterfactual property substitutions, separating a correct answer from genuine use of the supplied evidence. Across seven models, accuracy on a property channel does not track how strongly a model responds when that property is replaced, and model performance does not always match chemists' intuition about which properties are easier.

MolP2SBench recasts molecular property-to-structure inference as masked-scaffold multiple choice and adds shuffled and counterfactual property substitutions, separating a correct answer from genuine use of the supplied evidence. Across seven models, accuracy on a property channel does not track how strongly a model responds when that property is replaced, and model performance does not always match chemists' intuition about which properties are easier.

The work builds a unified benchmark for data-driven post-wildfire debris flow (PFDF) prediction, uses an LLM to convert operational weather forecast narratives into structured meteorological indicators and augments features with terrain and soil-moisture context, and proposes a reinforcement-learning feature-selection framework that identifies key drivers by perturbing features until positive and negative events become indistinguishable; learned models consistently outperform the fixed-coefficient USGS logistic-regression baseline, augmented features further improve most model families, and storm variables dominate globally while feature importance varies substantially across regions.

The work builds a unified benchmark for data-driven post-wildfire debris flow (PFDF) prediction, uses an LLM to convert operational weather forecast narratives into structured meteorological indicators and augments features with terrain and soil-moisture context, and proposes a reinforcement-learning feature-selection framework that identifies key drivers by perturbing features until positive and negative events become indistinguishable; learned models consistently outperform the fixed-coefficient USGS logistic-regression baseline, augmented features further improve most model families, and storm variables dominate globally while feature importance varies substantially across regions.

The authors introduce software-in-the-loop reconstruction (SWR), a self-supervised framework that draws reference outputs and verification targets from existing software workflows, instantiated with 500 workflows across 46 software families and six domains, where Qwen3.8-Max solves 838 tasks over three attempts and yields 1,422 verified trajectories oversampled into 3,000 reconstruction-only training examples, and supervised fine-tuning raises Qwen3.8-27B mean Terminal-Bench 2 performance from 47.94% to 53.56% while achieving the highest mean among four matched-token corpus controls on all four reported evaluations.

The authors introduce software-in-the-loop reconstruction (SWR), a self-supervised framework that draws reference outputs and verification targets from existing software workflows, instantiated with 500 workflows across 46 software families and six domains, where Qwen3.8-Max solves 838 tasks over three attempts and yields 1,422 verified trajectories oversampled into 3,000 reconstruction-only training examples, and supervised fine-tuning raises Qwen3.8-27B mean Terminal-Bench 2 performance from 47.94% to 53.56% while achieving the highest mean among four matched-token corpus controls on all four reported evaluations.

BitNest introduces a bit-nested speculative decoding framework that embeds a low-precision draft directly into the higher-precision target representation and recovers the target through residual refinement, so draft and target share a single physical weight representation, with the progressive-precision design extended to the KV cache for long-context inference; across multiple 7B–8B edge-friendly LLMs and diverse workloads it reports an average speculative acceptance rate of 95.2% while closely preserving higher-precision model quality, and 1.48–1.61x end-to-end speedup over FP16 autoregressive decoding, with consistently competitive or higher decoding speedup on the LLaMA models supported by all representative self-speculative baselines.

BitNest introduces a bit-nested speculative decoding framework that embeds a low-precision draft directly into the higher-precision target representation and recovers the target through residual refinement, so draft and target share a single physical weight representation, with the progressive-precision design extended to the KV cache for long-context inference; across multiple 7B–8B edge-friendly LLMs and diverse workloads it reports an average speculative acceptance rate of 95.2% while closely preserving higher-precision model quality, and 1.48–1.61x end-to-end speedup over FP16 autoregressive decoding, with consistently competitive or higher decoding speedup on the LLaMA models supported by all representative self-speculative baselines.

Agent security detection has worked almost entirely at the application telemetry layer, namely the served tool manifest, the user prompt and the model's messages, yet some threats slip past that boundary. Researchers paired application-level agent telemetry with kernel-level syscall traces to build ACE, a paired-session corpus of 4,047 sessions and 17 threat models spanning six delivery-vector families and 14 of the 25 OWASP LLM and agentic threat categories. Across four detector families, kernel evidence is discriminative on its own, composing it with application-layer evidence generally outperforms either single-layer view, and the result generalizes to unseen attack families and transfers to an alternate agent runtime.

Agent security detection has worked almost entirely at the application telemetry layer, namely the served tool manifest, the user prompt and the model's messages, yet some threats slip past that boundary. Researchers paired application-level agent telemetry with kernel-level syscall traces to build ACE, a paired-session corpus of 4,047 sessions and 17 threat models spanning six delivery-vector families and 14 of the 25 OWASP LLM and agentic threat categories. Across four detector families, kernel evidence is discriminative on its own, composing it with application-layer evidence generally outperforms either single-layer view, and the result generalizes to unseen attack families and transfers to an alternate agent runtime.