Optimization
25 notes, newest first.
Turbo Harness: Instance-Adaptive Harness Optimization
The preprint introduces Turbo Harness, which reuses artifacts from a completed global harness-optimization run, summarizes them into a structured playbook, and trains a harness editor to generate instance-specific…
Score Centering Stabilizes Off-policy Reinforcement Learning
The preprint attributes instability in large-language-model reinforcement learning under training-inference mismatch to persistent drift between engines. It derives an additive score-centering correction that cancels…
ClawGym II: Exploring Black-Box RL on Agent Harness
Researchers present a new framework for reinforcement learning (RL) using complex agent harnesses. This framework is designed to optimize RL more stably and efficiently for long-horizon tasks. The researchers' method…
GoalEvolve: From Handcrafted Algorithm Priors to Goal-Driven Evolution of Physical Design Algorithms
A new framework called GoalEvolve is presented that aims to improve physical design algorithms for electronic circuits. The framework uses a goal-driven approach to identify and address unmet requirements in the…
Beyond Trial-and-Error: Agentic Optimization for Image-to-Video Adherence
Researchers have developed a new framework called 'Agentic Self-Improvement' to improve the performance of Image-to-Video (I2V) models, which are commonly used for automated content creation. The new framework aims to…
Preference Tree Optimization: Enhancing Goal-Oriented Dialogue with Look-Ahead Simulations
The research presents a new framework called Preference Tree Optimization (PTO) for enhancing dialogue systems in specialized domains with limited data. It focuses on Motivational Interviewing (MI), a counseling…
ElasticBack: Stealthy Conditional Backdoor in LLM-Agent Skills via Coupled Trigger-Rule Optimization
The paper presents ElasticBack, a stealthy conditional backdoor that can compromise LLM agents by planting a rule and trigger in skills. The backdoor only activates when both the rule and the trigger appear together…
MicroEvo: Knowledge-Guided LLM Sampling for Efficient Microarchitecture Design Space Exploration
A new method called MicroEvo is proposed for efficient exploration of microarchitecture design spaces. This approach utilizes a combination of large language models (LLMs) and Monte Carlo Tree Search (MCTS) to optimize…
On-Policy and Off-Policy Learning for Large Action Spaces
The paper discusses policy learning in interactive systems with large action spaces, where agents observe contexts, select actions, and receive partial feedback. The research focuses on two paradigms: on-policy…
WIDE: Boosting Adaptive LLM Inference via Token-level Dynamic Width Pruning
Researchers introduce WIDE, a dynamic token-level pruning framework for improving the efficiency of large language models (LLMs). The framework allows each token to select attention and FFN channel groups dynamically…
Decentralized Online Riemannian Optimization for Strongly Geodesically Convex Functions
The authors study decentralized online optimization for specific types of functions (strongly geodesically convex) on Riemannian manifolds with certain curvature properties. They propose a new network-error analysis to…
Harness Engineering for LLM-Driven GPU Kernel Generation
The paper discusses a system for optimizing GPU kernels using large language models (LLMs). The system separates an evaluation harness from a profile-backed optimization controller to enforce compilation, correctness…
LLMs and Agentic AI Systems for Smart Grids: A Tutorial on Architectures and Applications
This paper discusses how large language models (LLMs) and agentic AI systems are being used in smart grids for tasks like forecasting, optimization, and control. The authors present a design principle called the…
SelectInfer: Selective Neuron Loading and Computation for On-Device LLMs
Researchers present SelectInfer, a neuron-level optimization framework for Large Language Models (LLMs) that can run efficiently on resource-constrained edge devices. It reduces memory footprint and computation by…
Explaining Process Control Optimisation Recommendations via GradientSHAP and Implicit Differentiation
Researchers have developed a method to explain optimization recommendations for industrial processes using GradientSHAP and Implicit Differentiation. This approach aims to bridge the trust gap between engineers who…
Selective Timestep Weighting and Advantage-Based Replay for Sample-Efficient Diffusion RLHF
The paper proposes two methods for improving the efficiency of Reinforcement Learning from Human Feedback (RLHF) when applied to diffusion models, which are often inefficient due to high feedback requirements. The…
Stability of Flow Models for Graph Signals
The paper presents research on the stability of continuous flow models for generating signals on graphs. The authors analyze normalized flow models parameterized by Graph Neural Networks (GNNs) and demonstrate that…
OptiAgent: End-to-End Optimization Modeling via Multi-Agent Iterative Refinement
The researchers have developed a multi-agent framework called OptiAgent that can take a natural language description of an Operations Research problem and produce a mathematically formulated solution along with…
OptiAgent: End-to-End Optimization Modeling via Multi-Agent Iterative Refinement
The authors propose OptiAgent, a multi-agent framework that takes a natural language description of an Operations Research problem and produces a solver-ready mathematical formulation and executable code. The…
TREK: Distill to Explore, Reinforce to Refine
The paper proposes a method called TREK (Teacher-Routed Exploration via Forward KL) to improve Group Relative Policy Optimization (GRPO). TREK uses distillation for exploration support expansion and can work with…
Understanding Agent-Based Patching of Compiler Missed Optimizations
The paper presents a study on the effectiveness of AI agents in addressing missed optimizations in compilers. The authors found that while coding agents can often optimize given examples, the generated patches…
Understanding Agent-Based Patching of Compiler Missed Optimizations
The paper 'Understanding Agent-Based Patching of Compiler Missed Optimizations' studies how well artificial intelligence (AI) agents can patch missed optimizations in compilers, where the compiler failed to optimize…
AutoMem: Automated Learning of Memory as a Cognitive Skill
Researchers developed AutoMem, a framework for improving a language model's memory management skills in long-horizon games. By treating memory as a trainable skill, they allowed the model to decide how to manage its…
Optimal Resource Utilization for Autonomous Laboratory Orchestrators
The authors propose a two-step method to optimize resource utilization in autonomous laboratories, specifically for metal-organic framework synthesis. The first step involves using constraint programming to find…
Parametric Open Source Games
The authors introduce a new concept called parametric open-source games, which are continuous versions of game equilibria where players choose parameter vectors to determine their actions in an underlying finite game…
