The Archive

Empirical evaluation finds tool graph memory rarely improves multi-turn agent tool-use over nine models, with gains costing 2x additional LLM calls.

Hardware verification framework uses LLM agents with independent data-integrity validation via scoreboards for coverage closure in RTL designs.

P³IBench introduces benchmark for proactive video-language models on perception, risk prediction, and intervention across 12 models in real-world streaming.

Theoretical framework derives sharp label-free calibration bounds for query-conditioned probabilistic predictors using minimax theorems and logical coherence.

BiShield-TEE identifies vulnerability in unilateral weight obfuscation for on-device LLMs via Solution Determinacy Property, exposing model structure.

Quantum machine learning content recommender using QNLP framework benchmarked against BiLSTM baseline on BBC iPlayer TV programme synopses.

DialFork benchmark measures procedural hallucination—fluent but logically incoherent agent behavior—via controlled dialogue fork injection.

Technique reduces qudit state preparation cost by learning sparse SNAP gate patterns to minimize phase parameters in bosonic quantum systems.

Benchmark comparing Muon optimizer vs AdamW across data/time/tuning budgets on 17.7M–33.6M LLMs and vision transformers, showing 3.25–22.44% validation-loss gains.

Proposes pairwise difference learning variant for classification with uncertainty quantification and meta-learning technique.

Controlled study of knowledge injection into addressed memory on GPT-2, Qwen3.8-Flash, and DeepSeek-V4.1-Flash examining transfer and behavior preservation.

Identifies harmful role of Classifier-Free Guidance in noisy phase of Text-Image-to-Image flow matching, proposes early-step correction for controllable image editing.

Analyzes Muon optimizer optimality via Random-Radius Isotropic Curvature Model, introduces geometric conditions for polar-factor update stability.

Primal-dual algorithm for reinforcement learning in partially observable constrained MDPs with finite-time convergence guarantees using finite-memory approximations.

Studies privacy leakage and data poisoning in federated fine-tuning of MobileNetV4-Conv-Small on Quick Draw with head-only FedAvg adaptation.

Survey of RL-driven diffusion models for language/vision generation, examining reward-based trajectory optimization analogous to RLHF for LLMs.

Lightweight HTTP/SSE-based Agent Status Protocol for heterogeneous multi-agent systems to report operational state and infer liveness via heartbeats.

Proposes Function-isolated Memory-Efficient Optimization (CiME) localizing low-overlap functional subnetworks in LLMs for continual learning without catastrophic forgetting.

Theoretical analysis of layer normalization placement effects on MLP approximation power using width augmentation and local linearization.

Framework for using LLMs as game masters in tabletop RPGs, addressing story generation, rule adherence, and player action freedom.

Hierarchical multitask audio model conditioning scene classification on event predictions, benchmarked across PaSST, PANNs, BEATs with information bottleneck analysis.

SCRIBE framework produces neuroscience-grounded interpretable semantic representations from LLM ensembles; validates context-dependent human semantic judgments with 921 participants.

Poison Valleys diagnostic traces path-dependent failures in LLM code repair agents via hypothesis/implementation tracking on 1,844 requests across PersistQueue and Dijkstra tasks.

QREF improves agent search via query refinement and evidence filtering to reduce misaligned queries and retrieval noise in iterative reasoning tasks.

Variance partitioning decomposes LLaMA2-7B brain alignment using SCRIBE's 65 neurocognitive semantic properties on Moth Radio Hour fMRI narratives.

ISW-Agent integrates LLM with n8n workflows for internal solitary wave remote sensing; achieves 94.2% task-match rate on 120 Chinese instructions.

Remote KV cache tier for GLM-5.3 MoE prefill parallelism shows context-parallel scaling bottleneck: p99 TTFT degrades 3.96s→17.3s with remote Valkey store.

OPRD uses multi-scope reminder supervision and outcome routing for long-horizon coding agents; combines autoregressive self-imitation with reminder-free student rollouts.

Phase kernels recover exponential storage capacity in Modern Hopfield Networks, addressing attractor basin overlap under high memory load.