Haicheng Chen

Shali Jiang, Hua Zheng, Boyang Liu et al.

Knowledge distillation (KD) transfers a single scalar prediction from a large foundation model (FM) to compact vertical models (VMs), suffering from diminishing transfer ratio -- the fraction of FM improvement captured by the VM -- as a single scalar cannot convey the rich intermediate knowledge that larger FMs learn. To address this bottleneck, we propose LoopFM (Learning frOm HistOrical ReP*resentations of FM), a framework that opens a high-bandwidth transfer channel by structuring FM intermediate embeddings as input features (e.g., user history sequence) for downstream VMs, without requiring real-time FM inference at serving and architectural coupling between FM and VM. We provide a theoretical framework for LoopFM with a gain decomposition and transfer-ratio analysis. On three public benchmarks, LoopFM demonstrates strong AUC improvements (e.g., 6\%+ on TaobaoAd) and complementary knowledge transfer capability with KD. On industrial-scale systems (billions of examples, trillion-parameter FMs), LoopFM approximately doubles the knowledge transfer ratio on top of KD, delivering a +0.5\% conversion improvement in Y1H1, and a +1.03\% and +1.22\% conversion improvement from two individual launches respectively in Y1H2.

Anssi Kanervisto, Stephanie Milani, Karolis Ramanauskas et al.

Reinforcement learning competitions advance the field by providing appropriate scope and support to develop solutions toward a specific problem. To promote the development of more broadly applicable methods, organizers need to enforce the use of general techniques, the use of sample-efficient methods, and the reproducibility of the results. While beneficial for the research community, these restrictions come at a cost -- increased difficulty. If the barrier for entry is too high, many potential participants are demoralized. With this in mind, we hosted the third edition of the MineRL ObtainDiamond competition, MineRL Diamond 2021, with a separate track in which we permitted any solution to promote the participation of newcomers. With this track and more extensive tutorials and support, we saw an increased number of submissions. The participants of this easier track were able to obtain a diamond, and the participants of the harder track progressed the generalizable solutions in the same task.

Max Taylor, Haicheng Chen, Feng Qin et al.

Control firmware in unmanned aerial vehicles (UAVs) uses sensors to model and manage flight operations, from takeoff to landing to flying between waypoints. However, sensors can fail at any time during a flight. If control firmware mishandles sensor failures, UAVs can crash, fly away, or suffer other unsafe conditions. In-situ model checking finds sensor failures that could lead to unsafe conditions by systematically failing sensors. However, the type of sensor failure and its timing within a flight affect its manifestation, creating a large search space. We propose Avis, an in-situ model checker to quickly uncover UAV sensor failures that lead to unsafe conditions. Widely used control firmware already support operating modes. Avis injects sensor failures as the control firmware transitions between modes - a key execution point where mishandled software exceptions can trigger unsafe conditions. We implemented Avis and applied it to ArduPilot and PX4. Avis found unsafe conditions 2.4X faster than Bayesian Fault Injection, the leading, state-of-the-art approach. Within the current code base of ArduPilot and PX4, Avis discovered 10 previously unknown software bugs that lead to unsafe conditions. Additionally, we reinserted 5 known bugs that caused serious, unsafe conditions and Avis correctly reported all of them.