arXiv subjects

Zhitao Zhu

Publications and source records attributed to Zhitao Zhu.

7 recordsLinked to original sources

arXiv2026

How Should LLMs Consume High-Quality Data? Optimal Data Scheduling via Quality-Aware Functional Scaling Laws

High-quality data is scarce in large language model (LLM) training, yet how to schedule its use with optimization dynamics lacks theoretical guidance. We extend functional scaling laws with time-varying data quality and derive asymptotically optimal joint data-quality and batch-size schedules within a feature-space regression model. The solution reveals two regimes and dual uses of high-quality data: in the noise-limited regime, a smaller batch converts cleaner data into more signal at comparable noise; in the signal-limited regime, late placement suppresses terminal noise without sacrificing signal accumulation. This explains why conventional decay schedules can conflict with curriculum-style pipelines. Motivated by the theoretical structure, we propose Drop-Stable-Rampup for LLM midtraining: drop the batch size at the quality transition, keep it low to accumulate signal, then ramp up to suppress noise. On a 15B MoE model midtrained on 108B tokens of general-domain proprietary data, Drop-Stable-Rampup improves average accuracy over Warmup-Stable-Decay by +1.70 and Cosine-decay by +2.98, including +4.23 on GSM8K and +2.80 on MATH. On a public math-and-code mixture, it leads all reported STEM, mathematics, and code benchmarks, improving the overall mean over the strongest baseline by +3.27 on a 600M dense model and +5.25 on the same MoE architecture.

cs.LG↗

arXiv2025

Provable Low-Rank Tensor-Train Approximations in the Inverse of Large-Scale Structured Matrices

This paper studies the low-rank property of the inverse of a class of large-scale structured matrices in the tensor-train (TT) format, which is typically discretized from differential operators. An interesting question that we are concerned with is: Does the inverse of the large-scale structured matrix still admit the low-rank TT representation with guaranteed accuracy? In this paper, we provide a computationally verifiable sufficient condition such that the inverse matrix can be well approximated in a low-rank TT format. It not only answers what kind of structured matrix whose inverse has the low-rank TT representation but also motivates us to develop an efficient TT-based method to compute the inverse matrix. Furthermore, we prove that the inverse matrix indeed has the low-rank tensor format for a class of large-scale structured matrices induced by differential operators involved in several PDEs, such as the Poisson, Boltzmann, and Fokker-Planck equations. Thus, the proposed algorithm is suitable for solving these PDEs with massive degrees of freedom. Numerical results on the Poisson, Boltzmann, and Fokker-Planck equations validate the correctness of our theory and the advantages of our methodology.

math.NA↗

arXiv2024

APTT: An accuracy-preserved tensor-train method for the Boltzmann-BGK equation

Solving the Boltzmann-BGK equation with traditional numerical methods suffers from high computational and memory costs due to the curse of dimensionality. In this paper, we propose a novel accuracy-preserved tensor-train (APTT) method to efficiently solve the Boltzmann-BGK equation. A second-order finite difference scheme is applied to discretize the Boltzmann-BGK equation, resulting in a tensor algebraic system at each time step. Based on the low-rank TT representation, the tensor algebraic system is then approximated as a TT-based low-rank system, which is efficiently solved using the TT-modified alternating least-squares (TT-MALS) solver. Thanks to the low-rank TT representation, the APTT method can significantly reduce the computational and memory costs compared to traditional numerical methods. Theoretical analysis demonstrates that the APTT method maintains the same convergence rate as that of the finite difference scheme. The convergence rate and efficiency of the APTT method are validated by several benchmark test cases.

math.NA↗

arXiv2022

Debias the Black-box: A Fair Ranking Framework via Knowledge Distillation

Deep neural networks can capture the intricate interaction history information between queries and documents, because of their many complicated nonlinear units, allowing them to provide correct search recommendations. However, service providers frequently face more complex obstacles in real-world circumstances, such as deployment cost constraints and fairness requirements. Knowledge distillation, which transfers the knowledge of a well-trained complex model (teacher) to a simple model (student), has been proposed to alleviate the former concern, but the best current distillation methods focus only on how to make the student model imitate the predictions of the teacher model. To better facilitate the application of deep models, we propose a fair information retrieval framework based on knowledge distillation. This framework can improve the exposure-based fairness of models while considerably decreasing model size. Our extensive experiments on three huge datasets show that our proposed framework can reduce the model size to a minimum of 1% of its original size while maintaining its black-box state. It also improves fairness performance by 15%~46% while keeping a high level of recommendation effectiveness.

cs.IR↗

arXiv2022

Leveraging Causal Inference for Explainable Automatic Program Repair

Deep learning models have made significant progress in automatic program repair. However, the black-box nature of these methods has restricted their practical applications. To address this challenge, this paper presents an interpretable approach for program repair based on sequence-to-sequence models with causal inference and our method is called CPR, short for causal program repair. Our CPR can generate explanations in the process of decision making, which consists of groups of causally related input-output tokens. Firstly, our method infers these relations by querying the model with inputs disturbed by data augmentation. Secondly, it generates a graph over tokens from the responses and solves a partitioning problem to select the most relevant components. The experiments on four programming languages (Java, C, Python, and JavaScript) show that CPR can generate causal graphs for reasonable interpretations and boost the performance of bug fixing in automatic program repair.

cs.SE↗

arXiv2022

Cali3F: Calibrated Fast Fair Federated Recommendation System

The increasingly stringent regulations on privacy protection have sparked interest in federated learning. As a distributed machine learning framework, it bridges isolated data islands by training a global model over devices while keeping data localized. Specific to recommendation systems, many federated recommendation algorithms have been proposed to realize the privacy-preserving collaborative recommendation. However, several constraints remain largely unexplored. One big concern is how to ensure fairness between participants of federated learning, that is, to maintain the uniformity of recommendation performance across devices. On the other hand, due to data heterogeneity and limited networks, additional challenges occur in the convergence speed. To address these problems, in this paper, we first propose a personalized federated recommendation system training algorithm to improve the recommendation performance fairness. Then we adopt a clustering-based aggregation method to accelerate the training process. Combining the two components, we proposed Cali3F, a calibrated fast and fair federated recommendation framework. Cali3F not only addresses the convergence problem by a within-cluster parameter sharing approach but also significantly boosts fairness by calibrating local models with the global model. We demonstrate the performance of Cali3F across standard benchmark datasets and explore the efficacy in comparison to traditional aggregation approaches.

cs.IR↗

arXiv2022

A Fair Federated Learning Framework With Reinforcement Learning

Federated learning (FL) is a paradigm where many clients collaboratively train a model under the coordination of a central server, while keeping the training data locally stored. However, heterogeneous data distributions over different clients remain a challenge to mainstream FL algorithms, which may cause slow convergence, overall performance degradation and unfairness of performance across clients. To address these problems, in this study we propose a reinforcement learning framework, called PG-FFL, which automatically learns a policy to assign aggregation weights to clients. Additionally, we propose to utilize Gini coefficient as the measure of fairness for FL. More importantly, we apply the Gini coefficient and validation accuracy of clients in each communication round to construct a reward function for the reinforcement learning. Our PG-FFL is also compatible to many existing FL algorithms. We conduct extensive experiments over diverse datasets to verify the effectiveness of our framework. The experimental results show that our framework can outperform baseline methods in terms of overall performance, fairness and convergence speed.

cs.LG↗