arXiv subjects

Tom Zhang

Publications and source records attributed to Tom Zhang.

6 recordsLinked to original sources

arXiv2026

MemMachine: A Ground-Truth-Preserving Memory System for Personalized AI Agents

Large Language Model (LLM) agents require persistent memory to maintain personalization, factual continuity, and long-horizon reasoning, yet standard context-window and retrieval-augmented generation (RAG) pipelines degrade over multi-session interactions. We present MemMachine, an open-source memory system that integrates short-term, long-term episodic, and profile memory within a ground-truth-preserving architecture that stores entire conversational episodes and reduces lossy LLM-based extraction. MemMachine uses contextualized retrieval that expands nucleus matches with surrounding context, improving recall when relevant evidence spans multiple dialogue turns. Across benchmarks, MemMachine achieves strong accuracy-efficiency tradeoffs: on LoCoMo it reaches 0.9169 using gpt4.1-mini; on LongMemEvalS (ICLR 2025), a six-dimension ablation yields 93.0 percent accuracy, with retrieval-stage optimizations -- retrieval depth tuning (+4.2 percent), context formatting (+2.0 percent), search prompt design (+1.8 percent), and query bias correction (+1.4 percent) -- outperforming ingestion-stage gains such as sentence chunking (+0.8 percent). GPT-5-mini exceeds GPT-5 by 2.6 percent when paired with optimized prompts, making it the most cost-efficient setup. Compared to Mem0, MemMachine uses roughly 80 percent fewer input tokens under matched conditions. A companion Retrieval Agent adaptively routes queries among direct retrieval, parallel decomposition, or iterative chain-of-query strategies, achieving 93.2 percent on HotpotQA-hard and 92.6 percent on WikiMultiHop under randomized-noise conditions. These results show that preserving episodic ground truth while layering adaptive retrieval yields robust, efficient long-term memory for personalized LLM agents.

cs.AI↗

arXiv2023

Optimizing and Fine-tuning Large Language Model for Urban Renewal

This study aims to innovatively explore adaptive applications of large language models (LLM) in urban renewal. It also aims to improve its performance and text generation quality for knowledge question-answering (QA) tasks. Based on the ChatGLM, we automatically generate QA datasets using urban renewal scientific literature corpora in a self-instruct manner and then conduct joint fine-tuning training on the model using the Prefix and LoRA fine-tuning methods to create an LLM for urban renewal. By guiding the LLM to automatically generate QA data based on prompt words and given text, it is possible to quickly obtain datasets in the urban renewal field and provide data support for the fine-tuning training of LLMs. The experimental results show that the joint fine-tuning training method proposed in this study can significantly improve the performance of LLM on the QA tasks. Compared with LoRA fine-tuning, the method improves the Bleu and Rouge metrics on the test by about 5%; compared with the model before fine-tuning, the method improves the Bleu and Rouge metrics by about 15%-20%. This study demonstrates the effectiveness and superiority of the joint fine-tuning method using Prefix and LoRA for ChatGLM in the urban renewal knowledge QA tasks. It provides a new approach for fine-tuning LLMs on urban renewal-related tasks.

cs.CL↗

arXiv2023

Estimating random close packing in polydisperse and bidisperse hard spheres via an equilibrium model of crowding

We show that an analogy between crowding in fluid and jammed phases of hard spheres captures the density dependence of the kissing number for a family of numerically generated jammed states. We extend this analogy to jams of mixtures of hard spheres in $d=3$ dimensions, and thus obtain an estimate of the random close packing (RCP) volume fraction, $\phi_{\textrm{RCP}}$, as a function of size polydispersity. We first consider mixtures of particle sizes with discrete distributions. For binary systems, we show agreement between our predictions and simulations, using both our own and results reported in previous works, as well as agreement with recent experiments from the literature. We then apply our approach to systems with continuous polydispersity, using three different particle size distributions, namely the log-normal, Gamma, and truncated power-law distributions. In all cases, we observe agreement between our theoretical findings and numerical results up to rather large polydispersities for all particle size distributions, when using as reference our own simulations and results from the literature. In particular, we find $\phi_{\textrm{RCP}}$ to increase monotonically with the relative standard deviation, $s_{\sigma}$, of the distribution, and to saturate at a value that always remains below 1. A perturbative expansion yields a closed-form expression for $\phi_{\textrm{RCP}}$ that quantitatively captures a distribution-independent regime for $s_{\sigma} < 0.5$. Beyond that regime, we show that the gradual loss in agreement is tied to the growth of the skewness of size distributions.

cond-mat.soft↗

arXiv2021

A Multi-Level Trace Clustering Analysis Scheme for Measuring Students' Self-Regulated Learning Behavior in a Master-Based Online Learning Environment

The study introduces a new analysis scheme to analyze trace data and visualize students' self-regulated learning strategies in a mastery-based online learning modules platform. The pedagogical design of the platform resulted in fewer event types and less variability in student trace data. The current analysis scheme overcomes those challenges by conducting three levels of clustering analysis. On the event level, mixture-model fitting is employed to distinguish between abnormally short and normal assessment attempts and study events. On the module level, trace level clustering is performed with three different methods for generating distance metrics between traces, with the best performing output used in the next step. On the sequence level, trace level clustering is performed on top of module-level clusters to reveal students' change of learning strategy over time. We demonstrated that distance metrics generated based on learning theory produced better clustering results than pure data-driven or hybrid methods. The analysis showed that most students started the semester with productive learning strategies, but a significant fraction shifted to a multitude of less productive strategies in response to increasing content difficulty and stress. The observations could prompt instructors to rethink conventional course structure and implement interventions to improve self-regulation at optimal times.

physics.ed-ph↗

arXiv2021

Analyzing the Heterogeneous Impact of Remote Learning on Students' Ability to Stay on Track During the Pandemic

This study investigates how remote learning due to the COVID pandemic impacts students' ability to keep up with the pace of instruction in a university level physics course, with a focus on the heterogeneous impact of remote learning on different demographic groups. Student learning data is collected from 70 online learning modules assigned as both online homework and self-study material in both Fall 2020 and Spring 2020 semesters, with the first 41 modules being assigned before campus closure in Spring 2020. Students' ability to stay on track is measured by three data indicators: percentage of modules submitted before the due date, percentage of modules submitted early for extra credit, and percentage of modules properly engaged with. The student population is divided into two demographic groups according to each of the four demographic variables: Gender, Ethnicity, Transfer Status and First generation status. All three data indicators from both semesters are first compared between each pair of demographic groups within the same semester, and then compared within the same group between the two semesters. Exam and course scores are also compared between demographics groups in Fall 2020 and against scores from the Fall 2019 semester. The analysis revealed that remote learning significantly reduced on-time and early submission for first-generation students, but increased the frequency of early submission for transfer students. Under-represented minority students had higher percentage of engaged modules during the pandemic, while non-URM students had the same level of module engagement before and after the pandemic. These results suggest that pandemic remote learning had very different impact for students with different demographic background, and future instructional design should strive to provide more flexible options to meet the various needs for a diverse higher-ed student population.

physics.ed-ph↗

arXiv2021

Measuring the Impact of COVID-19 Induced Campus Closure on Student Self-Regulated Learning in Physics Online Learning Modules

This paper examines the impact of COVID-19 induced campus closure on university students' self-regulated learning behavior by analyzing click-stream data collected from student interactions with 70 online learning modules in a university physics course. To do so, we compared the trend of six types of actions related to the three phases of self-regulated learning before and after campus closures and between two semesters. We found that campus closure changed students' planning and goal setting strategies for completing the assignments, but didn't have a detectable impact on the outcome or the time of completion, nor did it change students' self-reflection behavior. The results suggest that most students still manage to complete assignments on time during the pandemic, while the design of online learning modules might have provided the flexibility and support for them to do so.

physics.ed-ph↗