A Provably Efficient Sample Collection Strategy for Reinforcement Learning
One of the challenges in online reinforcement learning (RL) is that the agent needs to trade off the exploration of the environment and the exploitation of the samples to optimize its behavior. Whether we optimize for regret, sample complexity, state-space coverage or model estimation, we need to strike a different exploration-exploitation trade-off. In this paper, we propose to tackle the exploration-exploitation problem following a decoupled approach composed of: 1) An "objective-specific" algorithm that (adaptively) prescribes how many samples to collect at which states, as if it has access to a generative model (i.e., a simulator of the environment); 2) An "objective-agnostic" sample collection exploration strategy responsible for generating the prescribed samples as fast as possible.
We would like to thank all the reviewers for your insightful and constructive reviews and your commendation for the
Reviewer #1: (1) Concerns on the quality of generated explanations. We will merge this in the revised version. We follow the dataset split used in [7] throughout all experiments. We will add the SoTA results for reference in the revised version. RE/ASC tasks so that annotated explanations are still representative.
Probabilistic Linear Solvers for Machine Learning
Linear systems are the bedrock of virtually all numerical computation. Machine learning poses specific challenges for the solution of such systems due to their scale, characteristic structure, stochasticity and the central role of uncertainty in the field. Unifying earlier work we propose a class of probabilistic linear solvers which jointly infer the matrix, its inverse and the solution from matrix-vector product observations. This class emerges from a fundamental set of desiderata which constrains the space of possible algorithms and recovers the method of conjugate gradients under certain conditions. We demonstrate how to incorporate prior spectral information in order to calibrate uncertainty and experimentally showcase the potential of such solvers for machine learning.
Malaysia downplays Huawei deal as U.S. checks China's AI reach
Malaysia declared it'll build a first-of-its-kind AI system powered by Huawei Technologies chips, only to distance itself from that statement a day later, underscoring the Asian nation's delicate position in the U.S.-Chinese AI race. Deputy Minister of Communications Teo Nie Ching said in a speech Monday her country would be the first to activate an unspecified class of Huawei "Ascend GPU-powered AI servers at national scale." Malaysia would deploy 3,000 units of Huawei's primary AI offering by 2026, she said in prepared remarks reviewed by Bloomberg News. Chinese startup DeepSeek would also make one of its AI models available to the Southeast Asian country, the official added.
Biden camp denies cancer was diagnosed earlier amid cover-up claims
Former United States President Joe Biden was not diagnosed with prostate cancer before last week, and received his "last known" blood test for the disease more than a decade ago, his office has said. The Biden camp's statement on Tuesday came as critics, including current President Donald Trump, stoked scepticism over the timing of the diagnosis, which has reanimated questions about whether the former president misled the public about his health while in office. "President Biden's last known PSA was in 2014," Biden's office said in the brief statement, referring to the prostate-specific antigen test used to detect prostate cancer. "Prior to Friday, President Biden had never been diagnosed with prostate cancer." On Monday, Trump said he was "surprised" that the public had not been notified about Biden's diagnosis "a long time ago".