arXiv subjects

Zhijian Li

Publications and source records attributed to Zhijian Li.

At least 19 recordsLinked to original sources

arXiv2026

Rebalancing Token Importance in Language Models with TF-IDF Weighted Cross-Entropy Loss

Large language models are typically trained under uniform token weighting, which allows frequent and low-information tokens to dominate learning and can increase the tendency to memorize surface-level text spans. To address this, we present an information-weighted cross-entropy loss that rescales token-level contributions using TF-IDF statistics, emphasizing semantically informative tokens while down-weighting ubiquitous ones. Experiments on five decoder-only LLMs ranging from 1.1B to 13B parameters show consistent reductions in memorized substring length while preserving perplexity and downstream task performance. Under LoRA fine-tuning, TF-IDF reduces average substring memorization length by 14% across all five models. Under full-weight fine-tuning on TinyLLaMA 1.1B, the reduction reaches 58%. Our approach is architecture-agnostic and can be incorporated into existing training pipelines with less than 3% computational overhead, offering a lightweight and principled way to mitigate memorization without disrupting standard training dynamics.

cs.CL↗

arXiv2026

Orbit-Planner: Towards Latent World Models for On-Orbit Obstacle Avoidance of Satellite Agents

Satellite agents for on-orbit navigation tasks need to predict collision risks using limited onboard observations. However, conventional planners often rely on predefined maps and fixed environmental assumptions, limiting their adaptability in dynamic on-orbit scenarios. In this paper, we propose Orbit-Planner, a two-stage latent world model for on-orbit obstacle avoidance. Orbit-Planner learns action-conditioned spacecraft dynamics to perform future-state rollouts in latent space, and introduces a Physics Probe to decode physical state changes from imagined latent trajectories. Experiments demonstrate that Orbit-Planner can perform long-horizon latent rollouts and recover physical states from imagined trajectories. In closed-loop obstacle-avoidance navigation in Isaac Sim, it attains a success rate of 91.7%. Code is available at https://github.com/ZhijianLi2003/Orbit_Planner.

cs.RO↗

arXiv2026

CoEvo-Mem: Co-Evolving Retrieval Policy and Memory Bank for LLM Agents

As memories accumulate across tasks and sessions, the performance of long-term LLM agents depends jointly on query-specific retrieval and continual memory refinement. However, existing methods typically optimize either memory access, through iterative query refinement or adaptive retrieval policies, or memory evolution such as structural update. This separation overlooks a fundamental feedback loop: retrieval determines which memories receive usage signals, while updated memory bank reshape future retrieval. We propose \textbf{CoEvo-Mem}, a closed-loop framework for co-evolving the retrieval policy and memory bank. For each query, a frozen LLM generates route-specific query rewrites and a routing prior, which a lightweight residual router corrects online. The retrieved context serves as the coupling interface between the two learning processes: task outcomes assign credit to routing decisions, while trajectory-conditioned feedback updates memory values and graph relations. These updates alter how memories are ranked and selected for subsequent queries, thereby closing the feedback loop. To mitigate coupling induced non-stationarity, CoEvo-Mem alternates between updating the router with the memory bank fixed and evolving the memory bank with the retrieval policy fixed. Across seven diverse benchmarks, \textbf{CoEvo-Mem} achieves state-of-the-art performance, demonstrating the importance of retrieval-memory coevolution.

cs.AI↗

arXiv2026

Exceptional-point-constrained locking of boundary-sensitive topological transitions in chiral non-Hermitian SSH-type lattices

Topological transitions in non-Hermitian systems are generally boundary sensitive: a point-gap winding transition under periodic boundary condition (PBC) and a non-Bloch bulk real-line-gap transition under open boundary condition (OBC) at $\mathrm{Re}(E)=0$ are governed by different spectra and therefore need not coincide. Here we show, for a class of chiral non-Hermitian Su--Schrieffer--Heeger (SSH)-type lattices, that these two criticalities can be locked by an exceptional-point-constrained (EP-constrained) parameter evolution. The key requirement is not the occurrence of isolated exceptional points, but the persistence of a zero-energy Bloch degeneracy along the entire sweep, which is generically exceptional in the non-Hermitian regime. In an analytically tractable limit of an extended non-Hermitian SSH chain, the EP-constrained manifolds and both transition boundaries are obtained in closed form, making the locking explicit. Away from this limit, numerical generalized-Brillouin-zone (GBZ) calculations confirm the correspondence for representative constrained sweeps, whereas unconstrained paths show that isolated exceptional points or Hermitian degeneracies do not enforce locking. We further verify the mechanism in a spinful four-band extension with branch-resolved GBZs, including strongly branch-imbalanced regimes. These results establish a path-dependent diagnostic principle: along EP-constrained sweeps in this SSH-type class, changes in PBC point-gap winding can indicate OBC non-Bloch bulk real-line-gap transitions and the corresponding changes in zero-energy boundary modes.

physics.optics↗

arXiv2026

ReasonSTL: Bridging Natural Language and Signal Temporal Logic via Tool-Augmented Process-Rewarded Learning

Signal Temporal Logic (STL) is an expressive formal language for specifying spatio-temporal requirements over real-valued, real-time signals. It has been widely used for the verification and synthesis of autonomous systems and cyber-physical systems. In practice, however, users often express their requirements in natural language rather than in structured STL formulas, making natural-language-to-STL translation a critical yet challenging task. Manual specification requires temporal-logic expertise and cannot scale, while prompting commercial LLM APIs incurs substantial token costs and may expose sensitive system requirements to third-party services, raising privacy concerns for industrial deployment. To address these challenges, we present \textsc{ReasonSTL}, a tool-augmented framework that adapts local open-source language models for natural-language-to-STL generation. \textsc{ReasonSTL} decomposes the translation process into explicit reasoning, deterministic tool calls, and structured formula construction. We further introduce process-rewarded training to supervise both tool-use trajectories and final formulas, together with \textsc{STL-Bench}, a bilingual, computation-aware benchmark grounded in real-world signals. Experiments show that a 4B model trained with \textsc{ReasonSTL} achieves state-of-the-art performance in both automatic metrics and human evaluations, demonstrating that \textsc{ReasonSTL} provides a transparent, low-cost, and privacy-preserving alternative for formal specification drafting.

cs.AI↗

arXiv2026

Anomalous Localization and Mobility Edges in Non-Hermitian Quasicrystals with Disordered Imaginary Gauge Fields

Localization in non-Hermitian quasicrystals can differ fundamentally from its Hermitian counterpart when non-reciprocity is spatially disordered. Here we study a one-dimensional non-Hermitian Aubry-André-Harper chain with a Bernoulli imaginary gauge field and quasiperiodic onsite modulation. In the nearest-neighbor limit, we identify an anomalous transition from a fully erratic non-Hermitian skin effect (ENHSE) phase to a fully localized phase. Although the fractal dimension vanishes in both regimes, the Lyapunov exponent and the fluctuation of the eigenstate center of mass sharply distinguish them. For generic finite-size realizations, this transition is further accompanied by a complex-to-real spectral change under periodic boundary conditions and a change of spectral winding from nontrivial to trivial. With weak next-nearest-neighbor hopping, we uncover an anomalous mobility edge at the same location as in the Hermitian generalized Aubry-André-Harper model, but separating Anderson-localized states from ENHSE-type macroscopic-accumulation states rather than extended states. We further show that this anomalous localization structure is reflected in spectral winding and wave-packet dynamics: single realizations exhibit winding-dependent drift, winding-resolved averaging preserves opposite directional responses, and full disorder averaging largely restores Hermitian-like transport. Our results establish practical diagnostics of anomalous localization and mobility edges in non-Hermitian quasicrystals.

cond-mat.dis-nn↗

arXiv2025

Topological Phase Transitions and Edge-State Transfer in Time-Multiplexed Quantum Walks

We investigate the topological phase transitions and edge-state properties of a time-multiplexed nonunitary quantum walk with sublattice symmetry. By constructing a Floquet operator incorporating tunable gain and loss, we systematically analyze both unitary and nonunitary regimes. In the unitary case, the conventional bulk-boundary correspondence (BBC) is preserved, with edge modes localized at opposite boundaries as predicted by topological invariants. In contrast, the nonunitary regime exhibits non-Hermitian skin effects, leading to a breakdown of the conventional BBC. By applying non-Bloch band theory and generalized Brillouin zones, we restore a generalized BBC and reveal a transfer phenomenon, where edge modes with different sublattice symmetries can become localized at the same boundary. Furthermore, we demonstrate that the structure of the spectral loops in the complex quasienergy plane provides a clear signature for these transfer behaviors. Our findings deepen the understanding of nonunitary topological phases and offer valuable insights for the experimental realization and control of edge states in non-Hermitian quantum systems.

quant-ph↗

arXiv2025

Kernel-Based Enhanced Oversampling Method for Imbalanced Classification

This paper introduces a novel oversampling technique designed to improve classification performance on imbalanced datasets. The proposed method enhances the traditional SMOTE algorithm by incorporating convex combination and kernel-based weighting to generate synthetic samples that better represent the minority class. Through experiments on multiple real-world datasets, we demonstrate that the new technique outperforms existing methods in terms of F1-score, G-mean, and AUC, providing a robust solution for handling imbalanced datasets in classification tasks.

cs.LG↗

arXiv2025

Document Classification using File Names

Rapid document classification is critical in several time-sensitive applications like digital forensics and large-scale media classification. Traditional approaches that rely on heavy-duty deep learning models fall short due to high inference times over vast input datasets and computational resources associated with analyzing whole documents. In this paper, we present a method using lightweight supervised learning models, combined with a TF-IDF feature extraction-based tokenization method, to accurately and efficiently classify documents based solely on file names, that substantially reduces inference time. Our results indicate that file name classifiers can process more than 90% of in-scope documents with 99.63% and 96.57% accuracy when tested on two datasets, while being 442x faster than more complex models such as DiT. Our method offers a crucial solution to efficiently process vast document datasets in critical scenarios, enabling fast and more reliable document classification.

cs.CL↗

arXiv2024

DeliLaw: A Chinese Legal Counselling System Based on a Large Language Model

Traditional legal retrieval systems designed to retrieve legal documents, statutes, precedents, and other legal information are unable to give satisfactory answers due to lack of semantic understanding of specific questions. Large Language Models (LLMs) have achieved excellent results in a variety of natural language processing tasks, which inspired us that we train a LLM in the legal domain to help legal retrieval. However, in the Chinese legal domain, due to the complexity of legal questions and the rigour of legal articles, there is no legal large model with satisfactory practical application yet. In this paper, we present DeliLaw, a Chinese legal counselling system based on a large language model. DeliLaw integrates a legal retrieval module and a case retrieval module to overcome the model hallucination. Users can consult professional legal questions, search for legal articles and relevant judgement cases, etc. on the DeliLaw system in a dialogue mode. In addition, DeliLaw supports the use of English for counseling. we provide the address of the system: https://data.delilegal.com/lawQuestion.

cs.CL↗

arXiv2024

Generating Hard-Negative Out-of-Scope Data with ChatGPT for Intent Classification

Intent classifiers must be able to distinguish when a user's utterance does not belong to any supported intent to avoid producing incorrect and unrelated system responses. Although out-of-scope (OOS) detection for intent classifiers has been studied, previous work has not yet studied changes in classifier performance against hard-negative out-of-scope utterances (i.e., inputs that share common features with in-scope data, but are actually out-of-scope). We present an automated technique to generate hard-negative OOS data using ChatGPT. We use our technique to build five new hard-negative OOS datasets, and evaluate each against three benchmark intent classifiers. We show that classifiers struggle to correctly identify hard-negative OOS utterances more than general OOS utterances. Finally, we show that incorporating hard-negative OOS data for training improves model robustness when detecting hard-negative OOS data and general OOS data. Our technique, datasets, and evaluation address an important void in the field, offering a straightforward and inexpensive way to collect hard-negative OOS data and improve intent classifiers' robustness.

cs.CL↗

arXiv2023

A Hybrid Method of Sentiment Analysis and Machine Learning Algorithm for the U.S. Presidential Election Forecasting

U.S. Presidential Election forecasting has been a research interest for several decades. Currently, election prediction consists of two main approaches: traditional models that incorporate economic data and poll surveys, and models that leverage Twitter (or X) and other social media platforms due to their increasing popularity in the past decade. However, traditional approaches have predominantly focused on national-level predictions, while social media-based approaches often oversimplify the nuanced differences between online discourse and the broader voting population's political landscape. In this work, we perform a hybrid method of both the machine learning algorithm and the sentiment analysis on the state level with various independent variables including census data, economic indicators, polling averages, and the newly defined average sentiment scores from Twitter. Our prediction for the 2020 U.S. Presidential Election yielded promising results. Most of our models successfully predicted a victory for the Democratic candidate with 96% accuracy using Gradient Boosting Trees and Multi-Layer Perceptron algorithms. This novel prediction framework addresses the limitations of existing U.S. Presidential Election forecasting approaches, particularly in terms of state-level predictions. It provides a valuable foundation for future research in this field and contributes to advancing our understanding of election dynamics.

cs.SI↗

arXiv2023

Feature Affinity Assisted Knowledge Distillation and Quantization of Deep Neural Networks on Label-Free Data

In this paper, we propose a feature affinity (FA) assisted knowledge distillation (KD) method to improve quantization-aware training of deep neural networks (DNN). The FA loss on intermediate feature maps of DNNs plays the role of teaching middle steps of a solution to a student instead of only giving final answers in the conventional KD where the loss acts on the network logits at the output level. Combining logit loss and FA loss, we found that the quantized student network receives stronger supervision than from the labeled ground-truth data. The resulting FAQD is capable of compressing model on label-free data, which brings immediate practical benefits as pre-trained teacher models are readily available and unlabeled data are abundant. In contrast, data labeling is often laborious and expensive. Finally, we propose a fast feature affinity (FFA) loss that accurately approximates FA loss with a lower order of computational complexity, which helps speed up training for high resolution image input.

cs.LG↗

arXiv2023

An Integrated Approach to Produce Robust Models with High Efficiency

Deep Neural Networks (DNNs) needs to be both efficient and robust for practical uses. Quantization and structure simplification are promising ways to adapt DNNs to mobile devices, and adversarial training is the most popular method to make DNNs robust. In this work, we try to obtain both features by applying a convergent relaxation quantization algorithm, Binary-Relax (BR), to a robust adversarial-trained model, ResNets Ensemble via Feynman-Kac Formalism (EnResNet). We also discover that high precision, such as ternary (tnn) and 4-bit, quantization will produce sparse DNNs. However, this sparsity is unstructured under advarsarial training. To solve the problems that adversarial training jeopardizes DNNs' accuracy on clean images and the struture of sparsity, we design a trade-off loss function that helps DNNs preserve their natural accuracy and improve the channel sparsity. With our trade-off loss function, we achieve both goals with no reduction of resistance under weak attacks and very minor reduction of resistance under strong attcks. Together with quantized EnResNet with trade-off loss function, we provide robust models that have high efficiency.

cs.CV↗

arXiv2022

Channel Pruning In Quantization-aware Training: An Adaptive Projection-gradient Descent-shrinkage-splitting Method

We propose an adaptive projection-gradient descent-shrinkage-splitting method (APGDSSM) to integrate penalty based channel pruning into quantization-aware training (QAT). APGDSSM concurrently searches weights in both the quantized subspace and the sparse subspace. APGDSSM uses shrinkage operator and a splitting technique to create sparse weights, as well as the Group Lasso penalty to push the weight sparsity into channel sparsity. In addition, we propose a novel complementary transformed l1 penalty to stabilize the training for extreme compression.

cs.LG↗

arXiv2022

An integrated recurrent neural network and regression model with spatial and climatic couplings for vector-borne disease dynamics

We developed an integrated recurrent neural network and nonlinear regression spatio-temporal model for vector-borne disease evolution. We take into account climate data and seasonality as external factors that correlate with disease transmitting insects (e.g. flies), also spill-over infections from neighboring regions surrounding a region of interest. The climate data is encoded to the model through a quadratic embedding scheme motivated by recommendation systems. The neighboring regions' influence is modeled by a long short-term memory neural network. The integrated model is trained by stochastic gradient descent and tested on leish-maniasis data in Sri Lanka from 2013-2018 where infection outbreaks occurred. Our model outperformed ARIMA models across a number of regions with high infections, and an associated ablation study renders support to our modeling hypothesis and ideas.

cs.LG↗

arXiv2021

DPVI: A Dynamic-Weight Particle-Based Variational Inference Framework

The recently developed Particle-based Variational Inference (ParVI) methods drive the empirical distribution of a set of \emph{fixed-weight} particles towards a given target distribution $π$ by iteratively updating particles' positions. However, the fixed weight restriction greatly confines the empirical distribution's approximation ability, especially when the particle number is limited. In this paper, we propose to dynamically adjust particles' weights according to a Fisher-Rao reaction flow. We develop a general Dynamic-weight Particle-based Variational Inference (DPVI) framework according to a novel continuous composite flow, which evolves the positions and weights of particles simultaneously. We show that the mean-field limit of our composite flow is actually a Wasserstein-Fisher-Rao gradient flow of certain dissimilarity functional $\mathcal{F}$, which leads to a faster decrease of $\mathcal{F}$ than the Wasserstein gradient flow underlying existing fixed-weight ParVIs. By using different finite-particle approximations in our general framework, we derive several efficient DPVI algorithms. The empirical results demonstrate the superiority of our derived DPVI algorithms over their fixed-weight counterparts.

cs.LG↗

arXiv2021

A Deep Learning Approach for Macroscopic Energy Consumption Prediction with Microscopic Quality for Electric Vehicles

This paper presents a machine learning approach to model the electric consumption of electric vehicles at macroscopic level, i.e., in the absence of a speed profile, while preserving microscopic level accuracy. For this work, we leveraged a high-performance, agent-based transportation tool to model trips that occur in the Greater Chicago region under various scenario changes, along with physics-based modeling and simulation tools to provide high-fidelity energy consumption values. The generated results constitute a very large dataset of vehicle-route energy outcomes that capture variability in vehicle and routing setting, and in which high-fidelity time series of vehicle speed dynamics is masked. We show that although all internal dynamics that affect energy consumption are masked, it is possible to learn aggregate-level energy consumption values quite accurately with a deep learning approach. When large-scale data is available, and with carefully tailored feature engineering, a well-designed model can overcome and retrieve latent information. This model has been deployed and integrated within POLARIS Transportation System Simulation Tool to support real-time behavioral transportation models for individual charging decision-making, and rerouting of electric vehicles.

cs.LG↗