arXiv subjects

Yekun Xu

Publications and source records attributed to Yekun Xu.

5 recordsLinked to original sources

arXiv2026

Towards Transparent RAG: Fostering Evidence Traceability in LLM Generation via Reinforcement Learning

Retrieval-Augmented Generation (RAG) delivers substantial value in knowledge-intensive applications. However, its generated responses often lack transparent reasoning paths that trace back to source evidence from retrieved documents. This opacity not only compromises the interpretability of the output but also limits the model's ability to fully exploit the provided context. To address this, we propose TRACE (Transparent RAG with evidenCE tracing), a framework designed to enhance evidence traceability in Large Language Models (LLMs) through reinforcement learning (RL). TRACE guides LLMs to produce structured outputs with explicit evidence citations by prompting and rewarding evidence relevance and proper formatting, alongside accuracy, to optimize structured traceability. To ensure training stability with multiple reward signals, we further introduce an adaptive strategy for merging rewards and adopt a stabilized KL-divergence estimator. Experiments on three multi-hop QA datasets using Qwen2.5-7B-Instruct and Llama-3.1-8B-Instruct show that TRACE achieves both transparent, evidence-attributed outputs and accuracy improvements of 10-30%. The resulting performance is comparable to advanced commercial LLMs (e.g., OpenAI o1, DeepSeek-R1). Further analyses demonstrate strong generalization capabilities to unseen tasks. Our code is publicly available now.

cs.CL↗

arXiv2021

A Generalization of a Theorem of Rothschild and van Lint

A classical result of Rothschild and van Lint asserts that if every non-zero Fourier coefficient of a Boolean function $f$ over $\mathbb{F}_2^{n}$ has the same absolute value, namely $|\hat{f}(α)|=1/2^k$ for every $α$ in the Fourier support of $f$, then $f$ must be the indicator function of some affine subspace of dimension $n-k$. In this paper we slightly generalize their result. Our main result shows that, roughly speaking, Boolean functions whose Fourier coefficients take values in the set $\{-2/2^k, -1/2^k, 0, 1/2^k, 2/2^k\}$ are indicator functions of two disjoint affine subspaces of dimension $n-k$ or four disjoint affine subspace of dimension $n-k-1$. Our main technical tools are results from additive combinatorics which offer tight bounds on the affine span size of a subset of $\mathbb{F}_2^{n}$ when the doubling constant of the subset is small.

cs.CC↗

arXiv2020

Quantum-Inspired Classical Algorithm for Slow Feature Analysis

Recently, there has been a surge of interest for quantum computation for its ability to exponentially speed up algorithms, including machine learning algorithms. However, Tang suggested that the exponential speed up can also be done on a classical computer. In this paper, we proposed an algorithm for slow feature analysis, a machine learning algorithm that extracts the slow-varying features, with a run time O(polylog(n)poly(d)). To achieve this, we assumed necessary preprocessing of the input data as well as the existence of a data structure supporting a particular sampling scheme. The analysis of algorithm borrowed results from matrix perturbation theory, which was crucial for the algorithm's correctness. This work demonstrates the possible application and extent for which quantum-inspired computation can be used.

cs.DM↗

arXiv2020

Quantum-Inspired Classical Algorithm for Principal Component Regression

This paper presents a sublinear classical algorithm for principal component regression. The algorithm uses quantum-inspired linear algebra, an idea developed by Tang. Using this technique, her algorithm for recommendation systems achieved runtime only polynomially slower than its quantum counterpart. Her work was quickly adapted to solve many other problems in sublinear time complexity. In this work, we developed an algorithm for principal component regression that runs in time polylogarithmic to the number of data points, an exponential speed up over the state-of-the-art algorithm, under the mild assumption that the input is given in some data structure that supports a norm-based sampling procedure. This exponential speed up allows for potential applications in much larger data sets.

cs.LG↗

arXiv2019

A New Algorithm for Finding Closest Pair of Vectors

Given $n$ vectors $x_0, x_1, \ldots, x_{n-1}$ in $\{0,1\}^{m}$, how to find two vectors whose pairwise Hamming distance is minimum? This problem is known as the \emph{Closest Pair Problem}. If these vectors are generated uniformly at random except two of them are correlated with Pearson-correlation coefficient $ρ$, then the problem is called the \emph{Light Bulb Problem}. In this work, we propose a novel coding-based scheme for the Closest Pair Problem. We design both randomized and deterministic algorithms, which achieve the best-known running time when the length of input vectors $m$ is small and the minimum distance is very small compared to $m$. Specifically, the running time of our randomized algorithm is $O(n\log^{2}n\cdot 2^{c m} \cdot \mathrm{poly}(m))$ and the running time of our deterministic algorithm is $O(n\log{n}\cdot 2^{c' m} \cdot \mathrm{poly}(m))$, where $c$ and $c'$ are constants depending only on the (relative) distance of the closest pair. When applied to the Light Bulb Problem, our result yields state-of-the-art deterministic running time when the Pearson-correlation coefficient $ρ$ is very large. Specifically, when $ρ\geq 0.9933$, our deterministic algorithm runs faster than the previously best deterministic algorithm (Alman, SOSA 2019).

cs.DS↗