arXiv subjects

Junhan Chen

Publications and source records attributed to Junhan Chen.

9 recordsLinked to original sources

arXiv2026

FlexiGrad: Adaptive Gradient Modulation for Hierarchical Fine-Grained Classification

Many fine-grained recognition tasks contain hierarchical labels such as order, family and species. Although this supervision should be beneficial, jointly optimising all levels often leads to unstable training because coarse and fine classifiers impose inconsistent gradients on the shared backbone. This hierarchical gradient conflict prevents the model from learning a coherent coarse-to-fine representation. In this paper, we propose FlexiGrad, a simple and parameter-free method that regulates gradient interactions during backpropagation. FlexiGrad removes only the harmful conflicting component when tasks disagree and reinforces the shared direction when they partially agree through a smooth hierarchy-aware weighting function. This produces stable optimisation and preserves both global structure and fine-grained discriminative cues. FlexiGrad integrates into existing architectures without modification while improves multi-granularity accuracy on CUB-200-2011, FGVC-Aircraft and Stanford Cars. The code will be available at PRIS-CV/FlexiGrad.

cs.CV↗

arXiv2026

Seeing as Experts Do: A Knowledge-Augmented Agent for Open-Set Fine-Grained Visual Understanding

Fine-grained visual understanding is shifting from static classification to knowledge-augmented reasoning, where models must justify as well as recognise. Existing approaches remain limited by closed-set taxonomies and single-label prediction, leading to significant degradation under open-set or context-dependent conditions. We present the Knowledge-Augmented Fine-Grained Reasoning Agent (KFRA), a unified framework that transforms fine-grained perception into evidence-driven reasoning. KFRA operates through a three-stage closed reasoning loop that emulates expert analysis. It first performs open-vocabulary detection and web-scale retrieval to generate category hypotheses. It then conducts discriminative regions localisation by aligning textual knowledge with visual evidence through a global-to-local focusing mechanism. Finally, it integrates all multimodal evidence within a large multimodal model to perform interpretable reasoning. Unlike existing agents that treat retrieval and reasoning as independent processes, KFRA establishes a retrieval-grounding coupling that converts retrieved knowledge into spatially grounded evidence for verification. This design enables factual, interpretable, and task-agnostic reasoning across diverse fine-grained scenarios. To evaluate this capability, we construct FGExpertBench, a benchmark designed to assess reasoning depth and cross-task generalisation across six knowledge dimensions. Extensive experiments demonstrate that KFRA consistently surpasses both standalone large multimodal models and current agent frameworks, achieving up to 19 percent improvement in reasoning accuracy and delivering evidence-grounded interpretability in open-set fine-grained visual understanding.

cs.CV↗

arXiv2025

Implicit Federated In-context Learning For Task-Specific LLM Fine-Tuning

As large language models continue to develop and expand, the extensive public data they rely on faces the risk of depletion. Consequently, leveraging private data within organizations to enhance the performance of large models has emerged as a key challenge. The federated learning paradigm, combined with model fine-tuning techniques, effectively reduces the number of trainable parameters. However,the necessity to process high-dimensional feature spaces results in substantial overall computational overhead. To address this issue, we propose the Implicit Federated In-Context Learning (IFed-ICL) framework. IFed-ICL draws inspiration from federated learning to establish a novel distributed collaborative paradigm, by converting client local context examples into implicit vector representations, it enables distributed collaborative computation during the inference phase and injects model residual streams to enhance model performance. Experiments demonstrate that our proposed method achieves outstanding performance across multiple text classification tasks. Compared to traditional methods, IFed-ICL avoids the extensive parameter updates required by conventional fine-tuning methods while reducing data transmission and local computation at the client level in federated learning. This enables efficient distributed context learning using local private-domain data, significantly improving model performance on specific tasks.

cs.LG↗

arXiv2025

Impact of Environmental Factors on LoRa 2.4 GHz Time of Flight Ranging Outdoors

In WSN/IoT, node localization is essential to long-running applications for accurate environment monitoring and event detection, often covering a large area in the field. Due to the lower time resolution of typical WSN/IoT platforms (e.g., 1 microsecond on ESP32 platforms) and the jitters in timestamping, packet-level localization techniques cannot provide meter-level resolution. For high-precision localization as well as world-wide interoperability via 2.4-GHz ISM band, a new variant of LoRa, called LoRa 2.4 GHz, was proposed by semtech, which provides a radio frequency (RF) time of flight (ToF) ranging method for meter-level localization. However, the existing datasets reported in the literature are limited in their coverages and do not take into account varying environmental factors such as temperature and humidity. To address these issues, LoRa 2.4 GHz RF ToF ranging data was collected on a sports field at the XJTLU south campus, where three LoRa nodes logged samples of ranging with a LoRa base station, together with temperature and humidity, at reference points arranged as a 3x3 grid covering 400 square meter over three weeks and uploaded all measurement records to the base station equipped with an ESP32-based transceiver for machine and user communications. The results of a preliminary investigation based on a simple deep neural network (DNN) model demonstrate that the environmental factors, including the temperature and humidity, significantly affect the accuracy of ranging, which calls for advanced methods of compensating for the effects of environmental factors on LoRa RF ToF ranging outdoors.

cs.NI↗

arXiv2023

On the use of QM/MM Surface Hopping simulations to understand thermally-activated rare event nonadiabatic transitions in the condensed phase

We implement a rare-event sampling scheme for quantifying the rate of thermally-activated nonadiabatic transitions in the condensed phase. Our QM/MM methodology uses the recently developed INAQS package to interface between an elementary electronic structure package and a popular open-source molecular dynamics software (GROMACS) to simulate an electron transfer event between two stationary ions in a solution of acetonitrile solvent molecules. Nonadiabatic effects are implemented through a surface hopping scheme and our simulations allow further quantitative insight into the participation ratio of solvent and the effect of ion separation distance as far as facilitating electron transfer. We also demonstrate that the standard gas-phase approaches for treating frustrated hops and velocity reversal must be refined when working in the condensed phase with many degrees of freedom. The code and methodology developed here can be easily expanded upon and modified to incorporate other systems, and should provide a great deal of new insight into a wide variety of condensed phase nonadiabatic phenomena.

physics.chem-ph↗

arXiv2023

Nonadiabatic Potential Energy Surfaces For A Molecule on a Surface as Found by Constrained Complete Active Space Theory

In order to study electron-transfer mediated chemical processes on a metal surface, one requires not one but two potential energy surfaces (one ground state and one excited state) as in Marcus theory. In this letter, we report that a novel, dynamically-weighted, state-averaged constrained CASSCF(2,2) (DW-SA-cCASSCF(2,2)) can produce such surfaces for the Anderson Impurity model. Both ground and excited state potentials are smooth, they incorporate states with a charge transfer character, and the accuracy of the ground state surface can be verified for some model problems by renormalization group theory. Future development of gradients and nonadiabatic derivative couplings should allow for the study of non-adiabatic dynamics for molecules near metal surfaces.

physics.chem-ph↗

arXiv2022

Active Spaces and Non-Orthogonal Configuration Interaction Approaches for Investigating Molecules on Metal Surfaces

We test a set of multiconfigurational wavefunction approaches for calculating the ground state electron population for a two-site Anderson model representing a molecule on a metal surface. In particular, we compare (i) a Hartree Fock like wavefunction where frontier orbitals are allowed to be nonorthogonal versus (ii) a fully non-orthogonal configuration interaction wavefunction based on constrained Hartree-Fock states. We test both the strong and weak metal-molecule hybridization ($Γ$) limits as well as the strong and weak electron-electron repulsion (U) limits. We obtain accurate results as compared with exact numerical renormalization group (NRG) theory, recovering charge transfer states where appropriate. The current framework should open a path to run molecular non-adiabatic dynamics on metal surfaces.

physics.chem-ph↗

arXiv2022

On the Meaning of Berry Force For Unrestricted Systems Treated With Mean-Field Electronic Structure

We show that the Berry force as computed by an approximate, mean-field electronic structure can be meaningful if properly interpreted. In particular, for a model Hamiltonian representing a molecular system with an even number of electrons interacting via a two-body (Hubbard) interaction and a spin-orbit coupling, we show that a meaningful nonzero Berry force emerges whenever there is spin unrestriction--even though the Hamiltonian is real-valued and formally the on-diagonal single-surface Berry force must be zero. Moreover, if properly applied, this mean-field Berry force yields roughly the correct asymptotic motion for scattering through an avoided crossing. That being said, within the context of a ground-state calculation, several nuances do arise as far interpreting the Berry force correctly, and as a practical matter, the Berry force diverges near the Coulson-Fisher point (which can lead to numerical instabilities). We do not address magnetic fields here.

physics.chem-ph↗

arXiv2020

Electronic Structure for Multielectronic Molecules Near a Metal Surface

We analyze a model problem representing a multi-electronic molecule sitting on a metal surface. Working with a reduced configuration interaction Hamiltonian, we show that one can extract very accurate ground state wavefunctions as compared with the numerical renormalization group theory (NRG) -- even in the limit of weak metal-molecule coupling strength but strong intramolecular electron-electron repulsion. Moreover, we extract what appear to be meaningful excitation energies as well. Our findings should lay the groundwork for future {\em ab initio} studies of charge transfer processes and bond making/breaking processes on metal surfaces.

physics.chem-ph↗