AI research from July 2026 Page 7

The paper introduces MetaPerch, a bioacoustics foundation model that jointly trains on primary species identification and auxiliary metadata prediction tasks to improve generalization against domain shifts.

ShortOPD uses a dynamic distillation strategy to fix structural collapse in pruned LLMs by adjusting training rollouts based on model output quality.

The paper introduces the Harness Handbook, a behavior-centric documentation system that helps agents and developers navigate and modify large, complex agent codebases.

KnowAct-GUIClaw is a framework for autonomous GUI manipulation that utilizes a memory-driven, two-tier architecture to manage complex cross-platform tasks.

OvisOCR2 is a model designed to parse visually rich documents into structured Markdown in a single pass.

This paper identifies a problem where AI-generated evaluation data for language models can silently contain hidden errors, leading to misleading results, and proposes a mandatory manual check to prevent these issues.

GigaWorld-Policy-0.5 improves real-time robotic control by decoupling action generation from future video simulation using a specialized Mixture-of-Transformers architecture.

ProfMalPlus uses a multi-agent reasoning framework to detect malicious NPM packages by combining static code analysis with dynamic verification.

The paper introduces a hybrid variational autoencoder framework that integrates longitudinal tumor growth measurements with time-to-event outcomes using genomic data to improve predictive accuracy.

The paper uses the Lyapunov characteristic exponent as a reward signal to teach reinforcement learning agents how to stabilize an inverted pendulum with vertical motion.

The researchers introduced Cluster-based Sequential Feature Selection (CSFS), a wrapper method that reduces the computational cost of feature selection in renewable energy prediction pipelines.

VideoRAE replaces pixel-focused autoencoders with a system that maps video foundation model features into more efficient and semantically aware latent representations.

The researchers developed a security framework to protect the lifecycle of reusable LLM agent skills from creation through execution.

The researchers developed a method to assign rewards to intermediate steps in agent interactions to improve performance on long-horizon tasks.

The paper introduces RELAI-VCL, a regression-aware optimizer that prevents agent performance from dropping when learning new tasks.

This paper proposes a new method for penetration testing AI-enabled systems that focuses on violating operational objectives through AI-governed behavior rather than just compromising resources.

The paper introduces Lighthouse RL, a reinforcement learning method that uses strategic reset points to improve sample efficiency in analog circuit sizing.

The paper identifies how architecture choices like normalization placement and width expansion prevent gradient rank collapse in deep Transformer models at initialization.

The paper examines how developers integrate agentic coding tools into their workflows by analyzing pull request patterns across 2,361 GitHub repositories.

The authors implement a domain routing system that dispatches page images to specialized OCR models, enabling accurate text extraction across varied historical Manchu writing styles.