arXiv subjects

Weiqi Luo

Publications and source records attributed to Weiqi Luo.

At least 37 records · Page 2Linked to original sources

arXiv2025

Advancing Mathematical Reasoning in Language Models: The Impact of Problem-Solving Data, Data Synthesis Methods, and Training Stages

Mathematical reasoning remains a challenging area for large language models (LLMs), prompting the development of math-specific LLMs such as LLEMMA, DeepSeekMath, and Qwen2-Math, among others. These models typically follow a two-stage training paradigm: pre-training with math-related corpora and post-training with problem datasets for supervised fine-tuning (SFT). Despite these efforts, the improvements in mathematical reasoning achieved through continued pre-training (CPT) are often less significant compared to those obtained via SFT. This study addresses this discrepancy by exploring alternative strategies during the pre-training phase, focusing on the use of problem-solving data over general mathematical corpora. We investigate three primary research questions: (1) Can problem-solving data enhance the model's mathematical reasoning capabilities more effectively than general mathematical corpora during CPT? (2) Are synthetic data from the same source equally effective, and which synthesis methods are most efficient? (3) How do the capabilities developed from the same problem-solving data differ between the CPT and SFT stages, and what factors contribute to these differences? Our findings indicate that problem-solving data significantly enhances the model's mathematical capabilities compared to general mathematical corpora. We also identify effective data synthesis methods, demonstrating that the tutorship amplification synthesis method achieves the best performance. Furthermore, while SFT facilitates instruction-following abilities, it underperforms compared to CPT with the same data, which can be partially attributed to its poor learning capacity for more challenging problem-solving data. These insights provide valuable guidance for optimizing the mathematical reasoning capabilities of LLMs, culminating in our development of a powerful mathematical base model called MathGPT-8B.

cs.CL↗

arXiv2025

What Are Step-Level Reward Models Rewarding? Counterintuitive Findings from MCTS-Boosted Mathematical Reasoning

Step-level reward models (SRMs) can significantly enhance mathematical reasoning performance through process supervision or step-level preference alignment based on reinforcement learning. The performance of SRMs is pivotal, as they serve as critical guidelines, ensuring that each step in the reasoning process is aligned with desired outcomes. Recently, AlphaZero-like methods, where Monte Carlo Tree Search (MCTS) is employed for automatic step-level preference annotation, have proven particularly effective. However, the precise mechanisms behind the success of SRMs remain largely unexplored. To address this gap, this study delves into the counterintuitive aspects of SRMs, particularly focusing on MCTS-based approaches. Our findings reveal that the removal of natural language descriptions of thought processes has minimal impact on the efficacy of SRMs. Furthermore, we demonstrate that SRMs are adept at assessing the complex logical coherence present in mathematical language while having difficulty in natural language. These insights provide a nuanced understanding of the core elements that drive effective step-level reward modeling in mathematical reasoning. By shedding light on these mechanisms, this study offers valuable guidance for developing more efficient and streamlined SRMs, which can be achieved by focusing on the crucial parts of mathematical reasoning.

cs.AI↗

arXiv2024

Expediting and Elevating Large Language Model Reasoning via Hidden Chain-of-Thought Decoding

Large language models (LLMs) have demonstrated remarkable capabilities in tasks requiring reasoning and multi-step problem-solving through the use of chain-of-thought (CoT) prompting. However, generating the full CoT process results in significantly longer output sequences, leading to increased computational costs and latency during inference. To address this challenge, we propose a novel approach to compress the CoT process through semantic alignment, enabling more efficient decoding while preserving the benefits of CoT reasoning. Our method introduces an auxiliary CoT model that learns to generate and compress the full thought process into a compact special token representation semantically aligned with the original CoT output. This compressed representation is then integrated into the input of the Hidden Chain-of-Thought (HCoT) model. The training process follows a two-stage procedure: First, the CoT model is optimized to generate the compressed token representations aligned with the ground-truth CoT outputs using a contrastive loss. Subsequently, with the CoT model parameters frozen, the HCoT model is fine-tuned to generate accurate subsequent predictions conditioned on the prefix instruction and the compressed CoT representations from the CoT model. Extensive experiments across three challenging domains - mathematical reasoning, agent invocation, and question answering - demonstrate that our semantic compression approach achieves competitive or improved performance compared to the full CoT baseline, while providing significant speedups of at least 1.5x in decoding time. Moreover, incorporating contrastive learning objectives further enhances the quality of the compressed representations, leading to better CoT prompting and improved task accuracy. Our work paves the way for more efficient exploitation of multi-step reasoning capabilities in LLMs across a wide range of applications.

cs.CL↗

arXiv2024

Towards Robust Knowledge Tracing Models via k-Sparse Attention

Knowledge tracing (KT) is the problem of predicting students' future performance based on their historical interaction sequences. With the advanced capability of capturing contextual long-term dependency, attention mechanism becomes one of the essential components in many deep learning based KT (DLKT) models. In spite of the impressive performance achieved by these attentional DLKT models, many of them are often vulnerable to run the risk of overfitting, especially on small-scale educational datasets. Therefore, in this paper, we propose \textsc{sparseKT}, a simple yet effective framework to improve the robustness and generalization of the attention based DLKT approaches. Specifically, we incorporate a k-selection module to only pick items with the highest attention scores. We propose two sparsification heuristics : (1) soft-thresholding sparse attention and (2) top-$K$ sparse attention. We show that our \textsc{sparseKT} is able to help attentional KT models get rid of irrelevant student interactions and have comparable predictive performance when compared to 11 state-of-the-art KT models on three publicly available real-world educational datasets. To encourage reproducible research, we make our data and code publicly available at \url{https://github.com/pykt-team/pykt-toolkit}\footnote{We merged our model to the \textsc{pyKT} benchmark at \url{https://pykt.org/}.}.

cs.LG↗

arXiv2023

A Novel Residual-guided Learning Method for Image Steganography

Traditional steganographic techniques have often relied on manually crafted attributes related to image residuals. These methods demand a significant level of expertise and face challenges in integrating diverse image residual characteristics. In this paper, we introduce an innovative deep learning-based methodology that seamlessly integrates image residuals, residual distances, and image local variance to autonomously learn embedding probabilities. Our framework includes an embedding probability generator and three pivotal guiding components: Residual guidance strives to facilitate embedding in complex-textured areas. Residual distance guidance aims to minimize the residual differences between cover and stego images. Local variance guidance effectively safeguards against modifications in regions characterized by uncomplicated or uniform textures. The three components collectively guide the learning process, enhancing the security performance. Comprehensive experimental findings underscore the superiority of our approach when compared to traditional steganographic methods and randomly initialized ReLOAD in the spatial domain.

cs.CR↗

arXiv2023

AI-Generated Image Detection using a Cross-Attention Enhanced Dual-Stream Network

With the rapid evolution of AI Generated Content (AIGC), forged images produced through this technology are inherently more deceptive and require less human intervention compared to traditional Computer-generated Graphics (CG). However, owing to the disparities between CG and AIGC, conventional CG detection methods tend to be inadequate in identifying AIGC-produced images. To address this issue, our research concentrates on the text-to-image generation process in AIGC. Initially, we first assemble two text-to-image databases utilizing two distinct AI systems, DALLE2 and DreamStudio. Aiming to holistically capture the inherent anomalies produced by AIGC, we develope a robust dual-stream network comprised of a residual stream and a content stream. The former employs the Spatial Rich Model (SRM) to meticulously extract various texture information from images, while the latter seeks to capture additional forged traces in low frequency, thereby extracting complementary information that the residual stream may overlook. To enhance the information exchange between these two streams, we incorporate a cross multi-head attention mechanism. Numerous comparative experiments are performed on both databases, and the results show that our detection method consistently outperforms traditional CG detection techniques across a range of image resolutions. Moreover, our method exhibits superior performance through a series of robustness tests and cross-database experiments. When applied to widely recognized traditional CG benchmarks such as SPL2018 and DsTok, our approach significantly exceeds the capabilities of other existing methods in the field of CG detection.

cs.CV↗

arXiv2023

An Investigation of Darwiche and Pearl's Postulates for Iterated Belief Update

Belief revision and update, two significant types of belief change, both focus on how an agent modify her beliefs in presence of new information. The most striking difference between them is that the former studies the change of beliefs in a static world while the latter concentrates on a dynamically-changing world. The famous AGM and KM postulates were proposed to capture rational belief revision and update, respectively. However, both of them are too permissive to exclude some unreasonable changes in the iteration. In response to this weakness, the DP postulates and its extensions for iterated belief revision were presented. Furthermore, Rodrigues integrated these postulates in belief update. Unfortunately, his approach does not meet the basic requirement of iterated belief update. This paper is intended to solve this problem of Rodrigues's approach. Firstly, we present a modification of the original KM postulates based on belief states. Subsequently, we migrate several well-known postulates for iterated belief revision to iterated belief update. Moreover, we provide the exact semantic characterizations based on partial preorders for each of the proposed postulates. Finally, we analyze the compatibility between the above iterated postulates and the KM postulates for belief update.

cs.AI↗

arXiv2023

SC-Ques: A Sentence Completion Question Dataset for English as a Second Language Learners

Sentence completion (SC) questions present a sentence with one or more blanks that need to be filled in, three to five possible words or phrases as options. SC questions are widely used for students learning English as a Second Language (ESL). In this paper, we present a large-scale SC dataset, \textsc{SC-Ques}, which is made up of 289,148 ESL SC questions from real-world standardized English examinations. Furthermore, we build a comprehensive benchmark of automatically solving the SC questions by training the large-scale pre-trained language models on the proposed \textsc{SC-Ques} dataset. We conduct detailed analysis of the baseline models performance, limitations and trade-offs. The data and our code are available for research purposes from: \url{https://github.com/ai4ed/SC-Ques}.

cs.CL↗

arXiv2023

Improving Interpretability of Deep Sequential Knowledge Tracing Models with Question-centric Cognitive Representations

Knowledge tracing (KT) is a crucial technique to predict students' future performance by observing their historical learning processes. Due to the powerful representation ability of deep neural networks, remarkable progress has been made by using deep learning techniques to solve the KT problem. The majority of existing approaches rely on the \emph{homogeneous question} assumption that questions have equivalent contributions if they share the same set of knowledge components. Unfortunately, this assumption is inaccurate in real-world educational scenarios. Furthermore, it is very challenging to interpret the prediction results from the existing deep learning based KT models. Therefore, in this paper, we present QIKT, a question-centric interpretable KT model to address the above challenges. The proposed QIKT approach explicitly models students' knowledge state variations at a fine-grained level with question-sensitive cognitive representations that are jointly learned from a question-centric knowledge acquisition module and a question-centric problem solving module. Meanwhile, the QIKT utilizes an item response theory based prediction layer to generate interpretable prediction results. The proposed QIKT model is evaluated on three public real-world educational datasets. The results demonstrate that our approach is superior on the KT prediction task, and it outperforms a wide range of deep learning based KT models in terms of prediction accuracy with better model interpretability. To encourage reproducible results, we have provided all the datasets and code at \url{https://pykt.org/}.

cs.LG↗

arXiv2023

simpleKT: A Simple But Tough-to-Beat Baseline for Knowledge Tracing

Knowledge tracing (KT) is the problem of predicting students' future performance based on their historical interactions with intelligent tutoring systems. Recently, many works present lots of special methods for applying deep neural networks to KT from different perspectives like model architecture, adversarial augmentation and etc., which make the overall algorithm and system become more and more complex. Furthermore, due to the lack of standardized evaluation protocol \citep{liu2022pykt}, there is no widely agreed KT baselines and published experimental comparisons become inconsistent and self-contradictory, i.e., the reported AUC scores of DKT on ASSISTments2009 range from 0.721 to 0.821 \citep{minn2018deep,yeung2018addressing}. Therefore, in this paper, we provide a strong but simple baseline method to deal with the KT task named \textsc{simpleKT}. Inspired by the Rasch model in psychometrics, we explicitly model question-specific variations to capture the individual differences among questions covering the same set of knowledge components that are a generalization of terms of concepts or skills needed for learners to accomplish steps in a task or a problem. Furthermore, instead of using sophisticated representations to capture student forgetting behaviors, we use the ordinary dot-product attention function to extract the time-aware information embedded in the student learning interactions. Extensive experiments show that such a simple baseline is able to always rank top 3 in terms of AUC scores and achieve 57 wins, 3 ties and 16 loss against 12 DLKT baseline methods on 7 public datasets of different domains. We believe this work serves as a strong baseline for future KT research. Code is available at \url{https://github.com/pykt-team/pykt-toolkit}\footnote{We merged our model to the \textsc{pyKT} benchmark at \url{https://pykt.org/}.}.

cs.LG↗

arXiv2023

Enhancing Deep Knowledge Tracing with Auxiliary Tasks

Knowledge tracing (KT) is the problem of predicting students' future performance based on their historical interactions with intelligent tutoring systems. Recent studies have applied multiple types of deep neural networks to solve the KT problem. However, there are two important factors in real-world educational data that are not well represented. First, most existing works augment input representations with the co-occurrence matrix of questions and knowledge components\footnote{\label{ft:kc}A KC is a generalization of everyday terms like concept, principle, fact, or skill.} (KCs) but fail to explicitly integrate such intrinsic relations into the final response prediction task. Second, the individualized historical performance of students has not been well captured. In this paper, we proposed \emph{AT-DKT} to improve the prediction performance of the original deep knowledge tracing model with two auxiliary learning tasks, i.e., \emph{question tagging (QT) prediction task} and \emph{individualized prior knowledge (IK) prediction task}. Specifically, the QT task helps learn better question representations by predicting whether questions contain specific KCs. The IK task captures students' global historical performance by progressively predicting student-level prior knowledge that is hidden in students' historical learning interactions. We conduct comprehensive experiments on three real-world educational datasets and compare the proposed approach to both deep sequential KT models and non-sequential models. Experimental results show that \emph{AT-DKT} outperforms all sequential models with more than 0.9\% improvements of AUC for all datasets, and is almost the second best compared to non-sequential models. Furthermore, we conduct both ablation studies and quantitative analysis to show the effectiveness of auxiliary tasks and the superior prediction outcomes of \emph{AT-DKT}.

cs.CY↗

arXiv2023

pyKT: A Python Library to Benchmark Deep Learning based Knowledge Tracing Models

Knowledge tracing (KT) is the task of using students' historical learning interaction data to model their knowledge mastery over time so as to make predictions on their future interaction performance. Recently, remarkable progress has been made of using various deep learning techniques to solve the KT problem. However, the success behind deep learning based knowledge tracing (DLKT) approaches is still left somewhat unknown and proper measurement and analysis of these DLKT approaches remain a challenge. First, data preprocessing procedures in existing works are often private and custom, which limits experimental standardization. Furthermore, existing DLKT studies often differ in terms of the evaluation protocol and are far away real-world educational contexts. To address these problems, we introduce a comprehensive python based benchmark platform, \textsc{pyKT}, to guarantee valid comparisons across DLKT methods via thorough evaluations. The \textsc{pyKT} library consists of a standardized set of integrated data preprocessing procedures on 7 popular datasets across different domains, and 10 frequently compared DLKT model implementations for transparent experiments. Results from our fine-grained and rigorous empirical KT studies yield a set of observations and suggestions for effective DLKT, e.g., wrong evaluation setting may cause label leakage that generally leads to performance inflation; and the improvement of many DLKT approaches is minimal compared to the very first DLKT model proposed by Piech et al. \cite{piech2015deep}. We have open sourced \textsc{pyKT} and our experimental results at https://pykt.org/. We welcome contributions from other research groups and practitioners.

cs.LG↗

arXiv2022

DialogID: A Dialogic Instruction Dataset for Improving Teaching Effectiveness in Online Environments

Online dialogic instructions are a set of pedagogical instructions used in real-world online educational contexts to motivate students, help understand learning materials, and build effective study habits. In spite of the popularity and advantages of online learning, the education technology and educational data mining communities still suffer from the lack of large-scale, high-quality, and well-annotated teaching instruction datasets to study computational approaches to automatically detect online dialogic instructions and further improve the online teaching effectiveness. Therefore, in this paper, we present a dataset of online dialogic instruction detection, \textsc{DialogID}, which contains 30,431 effective dialogic instructions. These teaching instructions are well annotated into 8 categories. Furthermore, we utilize the prevalent pre-trained language models (PLMs) and propose a simple yet effective adversarial training learning paradigm to improve the quality and generalization of dialogic instruction detection. Extensive experiments demonstrate that our approach outperforms a wide range of baseline methods. The data and our code are available for research purposes from: https://github.com/ai4ed/DialogID.

cs.CL↗

arXiv2022

Wide & Deep Learning for Judging Student Performance in Online One-on-one Math Classes

In this paper, we investigate the opportunities of automating the judgment process in online one-on-one math classes. We build a Wide & Deep framework to learn fine-grained predictive representations from a limited amount of noisy classroom conversation data that perform better student judgments. We conducted experiments on the task of predicting students' levels of mastery of example questions and the results demonstrate the superiority and availability of our model in terms of various evaluation metrics.

cs.CL↗

arXiv2022

Dual Stream Computer-Generated Image Detection Network Based On Channel Joint And Softpool

With the development of computer graphics technology, the images synthesized by computer software become more and more closer to the photographs. While computer graphics technology brings us a grand visual feast in the field of games and movies, it may also be utilized by someone with bad intentions to guide public opinions and cause political crisis or social unrest. Therefore, how to distinguish the computer-generated graphics (CG) from the photographs (PG) has become an important topic in the field of digital image forensics. This paper proposes a dual stream convolutional neural network based on channel joint and softpool. The proposed network architecture includes a residual module for extracting image noise information and a joint channel information extraction module for capturing the shallow semantic information of image. In addition, we also design a residual structure to enhance feature extraction and reduce the loss of information in residual flow. The joint channel information extraction module can obtain the shallow semantic information of the input image which can be used as the information supplement block of the residual module. The whole network uses SoftPool to reduce the information loss of down-sampling for image. Finally, we fuse the two flows to get the classification results. Experiments on SPL2018 and DsTok show that the proposed method outperforms existing methods, especially on the DsTok dataset. For example, the performance of our model surpasses the state-of-the-art by a large margin of 3%.

cs.CV↗

arXiv2022

A Design of A Simple Yet Effective Exercise Recommendation System in K-12 Online Learning

We propose a simple but effective method to recommend exercises with high quality and diversity for students. Our method is made up of three key components: (1) candidate generation module; (2) diversity-promoting module; and (3) scope restriction module. The proposed method improves the overall recommendation performance in terms of recall, and increases the diversity of the recommended candidates by 0.81\% compared to the baselines.

cs.IR↗

arXiv2021

Universal Deep Network for Steganalysis of Color Image based on Channel Representation

Up to now, most existing steganalytic methods are designed for grayscale images, and they are not suitable for color images that are widely used in current social networks. In this paper, we design a universal color image steganalysis network (called UCNet) in spatial and JPEG domains. The proposed method includes preprocessing, convolutional, and classification modules. To preserve the steganographic artifacts in each color channel, in preprocessing module, we firstly separate the input image into three channels according to the corresponding embedding spaces (i.e. RGB for spatial steganography and YCbCr for JPEG steganography), and then extract the image residuals with 62 fixed high-pass filters, finally concatenate all truncated residuals for subsequent analysis rather than adding them together with normal convolution like existing CNN-based steganalyzers. To accelerate the network convergence and effectively reduce the number of parameters, in convolutional module, we carefully design three types of layers with different shortcut connections and group convolution structures to further learn high-level steganalytic features. In classification module, we employ a global average pooling and fully connected layer for classification. We conduct extensive experiments on ALASKA II to demonstrate that the proposed method can achieve state-of-the-art results compared with the modern CNN-based steganalyzers (e.g., SRNet and J-YeNet) in both spatial and JPEG domains, while keeping relatively few memory requirements and training time. Furthermore, we also provide necessary descriptions and many ablation experiments to verify the rationality of the network design.

cs.CV↗

arXiv2021

Improved Xception with Dual Attention Mechanism and Feature Fusion for Face Forgery Detection

With the rapid development of deep learning technology, more and more face forgeries by deepfake are widely spread on social media, causing serious social concern. Face forgery detection has become a research hotspot in recent years, and many related methods have been proposed until now. For those images with low quality and/or diverse sources, however, the detection performances of existing methods are still far from satisfactory. In this paper, we propose an improved Xception with dual attention mechanism and feature fusion for face forgery detection. Different from the middle flow in original Xception model, we try to catch different high-semantic features of the face images using different levels of convolution, and introduce the convolutional block attention module and feature fusion to refine and reorganize those high-semantic features. In the exit flow, we employ the self-attention mechanism and depthwise separable convolution to learn the global information and local information of the fused features separately to improve the classification the ability of the proposed model. Experimental results evaluated on three Deepfake datasets demonstrate that the proposed method outperforms Xception as well as other related methods both in effectiveness and generalization ability.

cs.CV↗