The Archive

Hessian-based automatic differentiation method for computing periodic orbits in dynamical systems, applied to double pendulum problem.

BUGSTONE-E2E framework mines CVE patch history to detect unknown vulnerabilities by extracting and generalizing code fixes from public databases.

Combines pruning, quantization, and distillation to compress Vision Transformers for crop disease detection on mobile agricultural devices.

Open toolkit for measuring context-dependent word sense representation in transformer embedding layers using controlled bridge-form constructs.

Embedded Graph Flows: permutation-equivariant generative model learning continuous embeddings for categorical graph node/edge generation.

LLM optimizes variable ordering heuristics for binary decision diagram-based reversible quantum circuit synthesis from Boolean functions.

Public-market scrutiny will intensify pressure on the Claude maker’s unusual attempt to balance profit and purpose.

It's the latest failure of OpenAI's internal monitoring and security systems.

Running reasoning and agentic AI at the edge has been harder than it needs to be. Until recently, models capable of multi-step reasoning were too large to run... Running reasoning and agentic AI at the edge has been harder than it needs to be. Until recently, models capable of multi-step reasoning were too large to run locally on edge hardware. Developers building agents have had to route inference through a data center, adding network dependency, increasing costs, and exposing data that may need to stay on device. That constraint is lifting. Source

RoboSPA benchmark evaluates Vision-Language-Action models on spatial reasoning and procedural complexity in robotic manipulation tasks.

AdaGate-DF proposes adaptive gated framework for deepfake detection optimized for low-resolution and resource-constrained environments.

Theoretical analysis of information aggregation rates in networked multi-agent learning with distributed feature access and DAG communication.

Systematic review of 66 studies on LLM applications for HVAC building automation, assessing deployment readiness and physical system safety.

Microsoft's Copilot rarely reproduces even full sentences from news articles and books, let alone substantive chunks that could substitute for the original, the company says in new legal filings as it fights copyright claims from publishers including The New York Times and book authors. As part of the lawsuit's discovery, Microsoft provided 8.2 million Copilot chat logs to an expert hired by news publishers. The logs, it claims, were specifically chosen "because they hit on keywords implicating use of News Plaintiffs' websites, and therefore the most likely to contain News Plaintiffs' works."...

It’s officially the Ternus era at Apple. Tim Cook stepped down as CEO this week, handing the company to former hardware chief John Ternus, whose first memo promised a “huge launch next week” — timing that puts Apple’s next iPhone event on his desk before he’s even settled in. Cook isn’t going far, though: he’s staying on as Executive Chairman, focused on the kind of policy […]

Mechanistic analysis of DeepSeek-V4-Flash's four-stream residual pathways reveals selective routing and distinct representation streams.

PPR algorithm detects environment changes in cooperative multi-agent RL using reward-derived signals for adaptive behavior coordination.

LexFlip benchmark with 373 Quebec French perturbations diagnoses legal meaning preservation in simplified clauses via dissociation method.

RISE method improves language model post-training via self-extrapolating policy distillation using model's own RLHF trajectory.

β-VAE ECG representations from ECGx.AI foundation model discriminate myocardial scarring patterns for cardiac diagnosis screening.

Framework for evaluating automatic NLG metrics via behavioral correctness assumptions under controlled response transformations.

GUT method quantifies reasoning uncertainty in LLMs by modeling branching complexity as graph structure, enabling detection of nonsensical reasoning chains.

Study tests interchangeability assumption in multi-agent LLM teams, finding role-matched agent swaps preserve task performance despite disruption.

Layer dropout re-evaluated for modern LLM pre-training; comprehensive study establishes conditions for improved efficiency without accuracy degradation.

Speculative Uncertainty method recovers failure signals for coding agents using draft-model cross-likelihoods, enabling safe execution without logit access.

AI4CDS framework guides computational design science with AI participation; case study on video safety surveillance illustrates human-AI responsibility division.

CONTINUITY framework ensures security-context consistency across composable LLM agent components via assume-guarantee contracts and verification.

Theoretical analysis of shallow neural network approximation in mixed Sobolev spaces with general activation functions.

Self-supervised contrastive learning framework for lexical representations enables large-scale phylogenetic inference without manual cognacy annotation.

Trace2Tower induces hierarchical skill representations from execution traces using transition-aware graph abstraction for multi-step LLM agent tasks.