Yimin Du

Yimin Du

Rolling-window factor pipelines for Chinese A-share markets contain a subtle but costly flaw: daily price-move limits (+/-10% main-board, +/-20% STAR/ChiNext) render a fraction of closing prices non-executable, yet standard implementations ingest these values before any row-filtering runs. The contaminated aggregates propagate silently through moving averages, correlations, and ranks--a failure mode we term "upstream contamination". On real A-share data it inflates apparent information coefficient by 18% while reducing realised Sharpe by 0.44 points, because the model learns to predict returns it cannot trade. We resolve this with a mask-first design: a Boolean tradability mask is constructed at data load time and threaded through every operator, so that no window ever reads a non-tradable price. Built on this foundation, the system adds (i) a GPU-vectorised 213-factor engine via PyTorch unfold primitives (51x over pandas); (ii) an Adjusted-MSE loss penalising wrong-sign predictions 11x more heavily than magnitude errors; (iii) block-bootstrap GBM augmentation; and (iv) Markowitz-Ledoit-Wolf portfolio optimisation with cvxpy warm-start caching. On a calibrated 3,000-stock synthetic panel the system achieves annualised Sharpe 2.05; on proprietary real A-share data (2022-2024) it achieves Sharpe 1.63. Ablation shows the mask contract is the single largest contributor (+0.44), exceeding any model or loss choice. The full implementation is released under MIT licence at https://github.com/initial-d/ml-quant-trading.

Liang Wen, Yunke Cai, Fenrui Xiao et al.

This paper introduces Light-R1, an open-source suite for training long reasoning models using reproducible and cost-effective methodology. Given the proprietary nature of data used in the DeepSeek-R1 series, we develop an alternative approach leveraging exclusively public data and models. Our curriculum training progressively increases data difficulty, combined with multi-staged post-training. Our Light-R1-32B model, trained from Qwen2.5-32B-Instruct, outperforms DeepSeek-R1-Distill-Qwen-32B in math reasoning. Experimental results show that this curriculum approach becomes more effective when distinct, diverse datasets are available for different training stages: fine-tuning DeepSeek-R1-Distilled models (pre-tuned by DeepSeek team on proprietary data) with 3,000 challenging examples from our curriculum dataset yielded state-of-the-art 7B and 14B models, while the 32B model, Light-R1-32B-DS performed comparably to QwQ-32B and DeepSeek-R1. Furthermore, we extend our work by applying GRPO on long reasoning models. Our final Light-R1-14B-DS achieves SOTA performance among 14B models in math, with AIME24 & 25 scores of 74.0 and 60.2 respectively, surpassing many 32B models and DeepSeek-R1-Distill-Llama-70B. Despite math-focused training, Light-R1-14B-DS demonstrates strong cross-domain generalization. Light-R1 represents a significant advancement in making sophisticated reasoning models more accessible and implementable in real-world applications. Our models, training data and code have been made available at https://github.com/Qihoo360/Light-R1.

Chonghan Liu, Yimin Du, Qi An et al.

Large language models frequently exhibit suboptimal performance on low resource languages, primarily due to inefficient subword segmentation and systemic training data imbalances. In this paper, we propose Variable Entropy Policy Optimization (VEPO), which leverages Reinforcement Learning with Verifiable Rewards to incorporate deterministic structural constraints into the policy alignment process. This framework ensures prescribed sequence length, robust format consistency, and rigorous linguistic well formedness, all enforced during training. Central to our approach is a variable entropy mechanism that enables the model to dynamically calibrate the equilibrium between literal fidelity and semantic naturalness by modulating the exploration exploitation manifold. By integrating entropy tempered advantage estimation with asymmetric clipping, VEPO sustains robust exploration while mitigating policy collapse. Empirical evaluations across 90 FLORES-200, COMET-22, chrF directions demonstrate that VEPO yields substantial improvements in both tokenization efficiency and translation quality, bridging the performance gap for underrepresented languages.

Yimin Du

FastText has established itself as a fundamental algorithm for learning word representations, demonstrating exceptional capability in handling out-of-vocabulary words through character-level n-gram embeddings. However, its hash-based bucketing mechanism introduces critical limitations for large-scale industrial deployment: hash collisions cause semantic drift, and memory requirements become prohibitively expensive when dealing with real-world vocabularies containing millions of terms. This paper presents a comprehensive memory optimization framework that fundamentally reimagines FastText's memory management through the integration of double-array trie (DA-trie) structures and mark-compact garbage collection principles. Our approach leverages the linguistic insight that n-grams sharing common prefixes or suffixes exhibit highly correlated embeddings due to co-occurrence patterns in natural language. By systematically identifying and merging semantically similar embeddings based on structural relationships, we achieve compression ratios of 4:1 to 10:1 while maintaining near-perfect embedding quality. The algorithm consists of four sophisticated phases: prefix trie construction with embedding mapping, prefix-based similarity compression, suffix-based similarity compression, and mark-compact memory reorganization. Comprehensive experiments on a 30-million Chinese vocabulary dataset demonstrate memory reduction from over 100GB to approximately 30GB with negligible performance degradation. Our industrial deployment results show significant cost reduction, faster loading times, and improved model reliability through the elimination of hash collision artifacts. Code and experimental implementations are available at: https://github.com/initial-d/me_fasttext

Zhongxuan Zhang, Bi Zeng, Xinyu Ni et al.

RGB-T tracking leverages the complementary strengths of RGB and thermal infrared (TIR) modalities to address challenging scenarios such as low illumination and adverse weather. However, existing methods often fail to effectively integrate temporal information and perform efficient cross-modal interactions, which constrain their adaptability to dynamic targets. In this paper, we propose BTMTrack, a novel framework for RGB-T tracking. The core of our approach lies in the dual-template backbone network and the Temporal-Modal Candidate Elimination (TMCE) strategy. The dual-template backbone effectively integrates temporal information, while the TMCE strategy focuses the model on target-relevant tokens by evaluating temporal and modal correlations, reducing computational overhead and avoiding irrelevant background noise. Building upon this foundation, we propose the Temporal Dual Template Bridging (TDTB) module, which facilitates precise cross-modal fusion through dynamically filtered tokens. This approach further strengthens the interaction between templates and the search region. Extensive experiments conducted on three benchmark datasets demonstrate the effectiveness of BTMTrack. Our method achieves state-of-the-art performance, with a 72.3% precision rate on the LasHeR test set and competitive results on RGBT210 and RGBT234 datasets.

Yimin Du

This paper presents a sophisticated multi-day turnover quantitative trading algorithm that integrates advanced deep learning techniques with comprehensive cross-sectional stock prediction for the Chinese A-share market. Our framework combines five interconnected modules: initial stock selection through deep cross-sectional prediction networks, opening signal distribution analysis using mixture models for arbitrage identification, market capitalization and liquidity-based dynamic position sizing, grid-search optimized profit-taking and stop-loss mechanisms, and multi-granularity volatility-based market timing models. The algorithm employs a novel approach to balance capital efficiency with risk management through adaptive holding periods and sophisticated entry/exit timing. Trained on comprehensive A-share data from 2010-2020 and rigorously backtested on 2021-2024 data, our method achieves remarkable performance with 15.2\% annualized returns, maximum drawdown constrained below 5\%, and a Sharpe ratio of 1.87. The strategy demonstrates exceptional scalability by maintaining 50-100 daily positions with a 9-day maximum holding period, incorporating dynamic profit-taking and stop-loss mechanisms that enhance capital turnover efficiency while preserving risk-adjusted returns. Our approach exhibits robust performance across various market regimes while maintaining high capital capacity suitable for institutional deployment.

Yimin Du

The quality of human preference data is crucial for training and evaluating large language models (LLMs), particularly in reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) scenarios. Traditional side-by-side (SBS) annotation approaches often struggle with inherent uncertainty, annotator disagreement, and the complexity of preference judgments. This paper introduces a novel framework based on intuitionistic fuzzy sets (IFS) for modeling and aggregating human preferences in LLM data annotation tasks. Our approach captures not only the degree of preference but also the uncertainty and hesitation inherent in human judgment through membership, non-membership, and hesitation degrees. We propose an IFS-based annotation protocol that enables more nuanced preference modeling, develops aggregation methods for handling annotator disagreement, and introduces quality metrics for preference data assessment. Experimental validation on multiple datasets demonstrates that our IFS-based approach significantly improves annotation consistency, reduces annotator fatigue, and produces higher-quality preference data compared to traditional binary and Likert-scale methods. The resulting preference datasets lead to improved model performance in downstream tasks, with 12.3\% improvement in win-rate against baseline models and 15.7\% reduction in annotation time. Our framework provides a principled approach to handling uncertainty in human preference annotation and offers practical benefits for large-scale LLM training.