Tag
Not Diamond released a methodology for model routing that achieves Opus-level quality while reducing agent costs by 20–80%, using a sequential decision approach to handle long-running coding agents effectively.
CDPR is a counterfactual advantage-based credit assignment method for training cost-aware sequential medical diagnosis models using reinforcement learning, improving accuracy while reducing examination costs and number.
GenMatch is an end-to-end generative matching framework for micro-view order-dispatching in ride-hailing, addressing challenges like batch encoding and utility learning to improve dispatch quality, with demonstrated effectiveness in real-world tests.
A new paper on arXiv introduces an open-source library called LLMRouter with over 16 router implementations and a benchmark xRouteBench, demonstrating that learned routers can outperform fixed-model baselines by 14.6%.
This paper formalizes the problem of when to invoke LLMs in streaming inference systems as a risk-based sequential stopping problem. It proves theoretical guarantees and empirically validates the framework on turbofan degradation data.
Presents an uncertainty-aware geosteering framework integrating particle filtering for probabilistic subsurface interpretation with reinforcement learning for sequential decision-making, evaluated on an industrial simulator.
This paper studies when and how a planner should supplement a pre-trained simulator with real experiments in sequential decision problems, proposing Fisher-SEP to minimize posterior variance of a target policy's value.