Haoyang Fang

Wenting Ye, Hongfei Yang, Shuai Zhao et al.

The substitute-based recommendation is widely used in E-commerce to provide better alternatives to customers. However, existing research typically uses the customer behavior signals like co-view and view-but-purchase-another to capture the substitute relationship. Despite its intuitive soundness, we find that such an approach might ignore the functionality and characteristics of products. In this paper, we adapt substitute recommendation into language matching problem by taking product title description as model input to consider product functionality. We design a new transformation method to de-noise the signals derived from production data. In addition, we consider multilingual support from the engineering point of view. Our proposed end-to-end transformer-based model achieves both successes from offline and online experiments. The proposed model has been deployed in a large-scale E-commerce website for 11 marketplaces in 6 languages. Our proposed model is demonstrated to increase revenue by 19% based on an online A/B experiment.

Zelin He, Haotian Lin, Boran Han et al.

Agentic reinforcement learning (RL) enables LLM agents to improve continuously from environment rewards, yet the resulting policies do not systematically accumulate reusable strategies that generalize across tasks. Modular skills can provide such reusable strategies, yet existing skill-augmented RL methods decouple skill creation from policy optimization, risking adopting skills that conflict with the evolving policy. Inspired by Anthropic's Skill Creator, we introduce ReSkill, an RL-in-the-loop skill creation framework that reconciles skill evolution with policy learning. ReSkill exploits the group-wise structure of GRPO to naturally embed three mechanisms with only marginal additional overhead: (1) an assertion-driven skill creator that diagnoses failures from past experience and proposes conditional, trigger-based skill revisions; (2) within-group rollout sampling that enables controlled comparison of skill versions, capturing which version best supports the policy's ongoing learning; and (3) Thompson Sampling with adaptive discounting to balance exploration and exploitation in skill version selection as the policy evolves. Across several domains, ReSkill consistently outperforms existing memory and skill-based RL methods, with the largest gains on unseen tasks. Analysis of the skill lifecycle shows skills being automatically created, tested, refined, and pruned as the policy improves, demonstrating reconciled skill-policy co-evolution.

Bingqing Song, Boran Han, Shuai Zhang et al. · amazon-science

Having an LLM that aligns with human preferences is essential for accommodating individual needs, such as maintaining writing style or generating specific topics of interest. The majority of current alignment methods rely on fine-tuning or prompting, which can be either costly or difficult to control. Model steering algorithms, which modify the model output by constructing specific steering directions, are typically easy to implement and optimization-free. However, their capabilities are typically limited to steering the model into one of the two directions (i.e., bidirectional steering), and there has been no theoretical understanding to guarantee their performance. In this work, we propose a theoretical framework to understand and quantify the model steering methods. Inspired by the framework, we propose a confident direction steering method (CONFST) that steers LLMs via modifying their activations at inference time. More specifically, CONFST builds a confident direction that is closely aligned with users' preferences, and this direction is then added to the activations of the LLMs to effectively steer the model output. Our approach offers three key advantages over popular bidirectional model steering methods: 1) It is more powerful, since multiple (i.e. more than two) users' preferences can be aligned simultaneously; 2) It is simple to implement, since there is no need to determine which layer to add the steering vector to; 3) No explicit user instruction is required. We validate our method on GPT-2 XL (1.5B), Mistral (7B) and Gemma-it (9B) models for tasks that require shifting the output of LLMs across various topics and styles, achieving superior performance over competing methods.

Zhenmei Shi, Haoyang Fang, Yu-Wing Tai et al.

Psychological studies have found that human visual tracking system involves learning, memory, and planning. Despite recent successes, not many works have focused on memory and planning in deep learning based tracking. We are thus interested in memory augmented network, where an external memory remembers the evolving appearance of the target (foreground) object without backpropagation for updating weights. Our Dual Augmented Memory Network (DAWN) is unique in remembering both target and background, and using an improved attention LSTM memory to guide the focus on memorized features. DAWN is effective in unsupervised tracking in handling total occlusion, severe motion blur, abrupt changes in target appearance, multiple object instances, and similar foreground and background features. We present extensive quantitative and qualitative experimental comparison with state-of-the-art methods including top contenders in recent VOT challenges. Notably, despite the straightforward implementation, DAWN is ranked third in both VOT2016 and VOT2017 challenges with excellent success rate among all VOT fast trackers running at fps > 10 in unsupervised tracking in both challenges. We propose DAWN-RPN, where we simply augment our memory and attention LSTM modules to the state-of-the-art SiamRPN, and report immediate performance gain, thus demonstrating DAWN can work well with and directly benefit other models to handle difficult cases as well.