arXiv subjects

Jiyu Chen

Publications and source records attributed to Jiyu Chen.

16 recordsLinked to original sources

arXiv2025

2D coherent spectroscopy signatures of exciton condensation in Ta$_2$NiSe$_5$

We show that the nonlinear optical response probed by two-dimensional coherent spectroscopy (2DCS) can discriminate between excitonic and lattice driven order. In the excitonic regime of a realistic model of Ta$_2$NiSe$_5$, the third order 2DCS signals are strongly enhanced by the condensate's amplitude and phase modes, with negligible contributions from single-particle excitations. In the linear optical response, in contrast, single-particle and collective-mode contributions overlap. With increasing electron-phonon coupling, the amplitude mode contribution to 2DCS initially remains robust, but then drops rapidly and remains small in the phonon-dominated regime -- even in systems with large order parameter. 2DCS also aids the detection of the massive relative phase mode, which is analogous to the Leggett mode in superconductors. Our analysis, based on the time-dependent Hartree-Fock approach, demonstrates that 2DCS can track the emergence of the symmetry-broken state and the crossover from Coulomb-driven to phonon-driven order.

cond-mat.str-el↗

arXiv2025

EdgeInfinite-Instruct: Bridging SFT-Based Optimization and NPU-Level Efficiency for Edge Devices

Deploying Transformer-based large language models (LLMs) on resource-constrained edge devices for long-sequence tasks remains challenging due to the quadratic time complexity of self-attention and growing Key-Value (KV) cache demands. While existing KV cache optimizations improve memory efficiency, they often fail to reduce time to first token (TTFT) and may degrade performance through token pruning. Alternative sequence modeling architectures address some of these limitations, but typically require full retraining and lack infrastructure support. EdgeInfinite offers an efficient solution by fine-tuning only a small subset of parameters, maintaining quality while reducing both computational and memory costs, including improved TTFT. However, its instruction-following ability is limited, and it lacks mobile-specific optimizations. To address these issues, we propose EdgeInfinite-Instruct, which introduces a Segmented Supervised Fine-Tuning (S-SFT) strategy tailored to long-sequence tasks such as summarization and question answering. We further optimized EdgeInfinite-Instruct for efficient deployment on edge NPUs by employing fine-grained post-training quantization (PTQ) to reduce computational demands while maintaining accuracy, and by implementing a fixed-shape computation graph that balances memory usage and on-device efficiency through scenario-specific customization of input token and cache sizes. Experiments on long-context benchmarks and real-world mobile tasks show that our approach improves domain-specific performance while maintaining efficiency on NPU-accelerated edge devices.

cs.CL↗

arXiv2025

CSIRO-LT at SemEval-2025 Task 11: Adapting LLMs for Emotion Recognition for Multiple Languages

Detecting emotions across different languages is challenging due to the varied and culturally nuanced ways of emotional expressions. The \textit{Semeval 2025 Task 11: Bridging the Gap in Text-Based emotion} shared task was organised to investigate emotion recognition across different languages. The goal of the task is to implement an emotion recogniser that can identify the basic emotional states that general third-party observers would attribute to an author based on their written text snippet, along with the intensity of those emotions. We report our investigation of various task-adaptation strategies for LLMs in emotion recognition. We show that the most effective method for this task is to fine-tune a pre-trained multilingual LLM with LoRA setting separately for each language.

cs.CL↗

arXiv2025

Amplitude mode in two-dimensional coherent spectroscopy of weak-coupling antiferromagnets

Two-dimensional coherent spectroscopy (2DCS) provides insights into the nonlinear response of correlated lattice systems. We simulate multipulse excitations in the Hubbard model using nonequilibrium dynamical mean-field theory to extract the 2DCS signal of weak-coupling antiferromagnets with and without local potential disorder. By comparing calculations with static and dynamic Hartree terms, and analyzing the waiting-time dependence of the signal, we identify the contribution of the collective amplitude mode to the spectroscopic features and the relevant underlying processes. With broadband pulses, the rephasing and nonrephasing peaks at the gap energy are found to be of predominant amplitude mode character. Using narrow-band pulses, we also demonstrate a strong enhancement of these amplitude mode-related signals at a pulse frequency of half the gap size.

cond-mat.str-el↗

arXiv2025

Ultrafast Electronic Structure Engineering in 1$T$-TaS$_2$: Role of Doping and Amplitude Mode Dynamics

In strongly correlated transition metal dichalcogenides, an intricate interplay of polaronic distortions, stacking arrangement, and electronic correlations determines the nature of the insulating state. Here, we study the response of the electronic structure to optical excitations to reveal the effect of chemical electron doping on this complex interplay. Transient changes in pristine and electron-doped 1$T$ -TaS$_2$ are measured by femtosecond time-resolved photoelectron spectroscopy and compared to theoretical modeling based on non-equilibrium dynamical mean-field theory and density functional theory. The fine changes in the oscillatory signal of the charge density wave amplitude mode indicate phase-dependent modifications in the Coulomb interaction and the hopping. Furthermore, we find an enhanced fraction of monolayers in the doped system. Our work demonstrates how the combination of time-resolved spectroscopy and advanced theoretical modeling provides insights into the physics of correlated transition metal dichalcogenides.

cond-mat.str-el↗

arXiv2025

EdgeInfinite: A Memory-Efficient Infinite-Context Transformer for Edge Devices

Transformer-based large language models (LLMs) encounter challenges in processing long sequences on edge devices due to the quadratic complexity of attention mechanisms and growing memory demands from Key-Value (KV) cache. Existing KV cache optimizations struggle with irreversible token eviction in long-output tasks, while alternative sequence modeling architectures prove costly to adopt within established Transformer infrastructure. We present EdgeInfinite, a memory-efficient solution for infinite contexts that integrates compressed memory into Transformer-based LLMs through a trainable memory-gating module. This approach maintains full compatibility with standard Transformer architectures, requiring fine-tuning only a small part of parameters, and enables selective activation of the memory-gating module for long and short context task routing. The experimental result shows that EdgeInfinite achieves comparable performance to baseline Transformer-based LLM on long context benchmarks while optimizing memory consumption and time to first token.

cs.CL↗

arXiv2024

Multidimensional coherent spectroscopy of correlated lattice systems

Multidimensional coherent spectroscopy (MDCS) has been established in quantum chemistry as a powerful tool for studying the nonlinear response and nonequilibrium dynamics of molecular systems. More recently, the technique has also been applied to correlated electron materials, where the interplay of localized and itinerant states makes the interpretation of the spectra more challenging. Here we use the Keldysh contour representation of effective models and nonequilibrium dynamical mean field theory to systematically study the MDCS signals of prototypical correlated lattice systems. By analyzing the current induced by sequences of ultrashort laser pulses we demonstrate the usefulness of MDCS as a diagnostic tool for excitation pathways and coherent processes in correlated solids. We also show that this technique allows to extract detailed information on the nature and evolution of photo-excited nonequilibrium states.

cond-mat.str-el↗

arXiv2024

Root-KGD: A Novel Framework for Root Cause Diagnosis Based on Knowledge Graph and Industrial Data

With the development of intelligent manufacturing and the increasing complexity of industrial production, root cause diagnosis has gradually become an important research direction in the field of industrial fault diagnosis. However, existing research methods struggle to effectively combine domain knowledge and industrial data, failing to provide accurate, online, and reliable root cause diagnosis results for industrial processes. To address these issues, a novel fault root cause diagnosis framework based on knowledge graph and industrial data, called Root-KGD, is proposed. Root-KGD uses the knowledge graph to represent domain knowledge and employs data-driven modeling to extract fault features from industrial data. It then combines the knowledge graph and data features to perform knowledge graph reasoning for root cause identification. The performance of the proposed method is validated using two industrial process cases, Tennessee Eastman Process (TEP) and Multiphase Flow Facility (MFF). Compared to existing methods, Root-KGD not only gives more accurate root cause variable diagnosis results but also provides interpretable fault-related information by locating faults to corresponding physical entities in knowledge graph (such as devices and streams). In addition, combined with its lightweight nature, Root-KGD is more effective in online industrial applications.

cs.AI↗

arXiv2024

Photo-induced insulator-metal transition in paramagnetic (V$_{1-x}$Cr$_{x}$)$_2$O$_3$

Pump-probe experiments with femtosecond time resolution allow to disentangle the electronic dynamics from the lattice response and thus provide valuable insights into the non-equilibrium behavior of correlated materials. In Cr-doped V$_2$O$_3$, a multi-orbital Mott-Hubbard material which has been intensively investigated for decades, time-resolved experiments reported a photo-induced insulator-metal transition leading to a transient metal state with nonthermal properties. Here, we combine non-equilibrium dynamical mean-field theory with realistic first principles modeling to simulate the ultrafast response of this material to a laser excitation. Our calculations reproduce the insulating initial state, with orbital occupations in agreement with experiment, and reveal an ultrafast pump-induced gap filling associated with a charge reshuffling between the $e_g^π$ and $a_{1g}$ orbitals. However, in contrast to the related compound VO$_2$, the electronic system thermalizes within a few tens of femtoseconds and we find no evidence for the existence of a metastable nonthermal metal. This suggests that the reported nonthermal behavior in the experiments may be associated with the mismatch between the electronic and lattice temperatures.

cond-mat.str-el↗

arXiv2023

Agents: An Open-source Framework for Autonomous Language Agents

Recent advances on large language models (LLMs) enable researchers and developers to build autonomous language agents that can automatically solve various tasks and interact with environments, humans, and other agents using natural language interfaces. We consider language agents as a promising direction towards artificial general intelligence and release Agents, an open-source library with the goal of opening up these advances to a wider non-specialist audience. Agents is carefully engineered to support important features including planning, memory, tool usage, multi-agent communication, and fine-grained symbolic control. Agents is user-friendly as it enables non-specialists to build, customize, test, tune, and deploy state-of-the-art autonomous language agents without much coding. The library is also research-friendly as its modularized design makes it easily extensible for researchers. Agents is available at https://github.com/aiwaves-cn/agents.

cs.CL↗

arXiv2023

Nature of the photo-induced metallic state in monoclinic VO$_2$

The metal-insulator transition of VO$_2$, which in equilibrium is associated with a structural phase transition, has been intensively studied for decades. In particular, it is challenging to disentangle the role of Mott physics from dimerization effects in the insulating phase. Femtosecond time-resolved experiments showed that optical excitations can induce a transient metallic state in the dimerized phase, which is distinct from the known equilibrium phases. In this study, we combine non-equilibrium cluster dynamical mean-field theory with realistic first principles modeling to clarify the nature of this laser-induced metallic state. We show that the doublon-holon production by laser pulses with polarization along the V-V dimers and the subsequent inter-orbital reshuffling of the photo-carriers leads to a population of orbital-mixed states and the filling of the gap. The photo-induced metal state is qualitatively similar to a hot electronic state in the dimerized structure, and does not involve a collapse of the Mott gap.

cond-mat.str-el↗

arXiv2023

Light-induced insulator-metal transition in Sr$_2$IrO$_4$ reveals the nature of the insulating ground state

Sr$_2$IrO$_4$ has attracted a lot of attention due to its structural and electronic similarities to La$_2$CuO$_4$ which is the parent compound of high-T$_c$ superconducting cuprates. It was proposed to be a strong spin-orbit coupled J$_{eff}$ = 1/2 Mott insulator, but the Mott nature of its insulating ground state and the origin of the gap have not been conclusively established. Here, we use ultrafast laser pulses to realize an insulator-metal transition in Sr$_2$IrO$_4$ and probe the resulting dynamics using time- and angle-resolved photoemission spectroscopy. We observe a closing of the gap and the formation of weakly-renormalized electronic bands in the gap region. Comparing these observations to the expected temperature and doping evolution of Mott gaps and Hubbard bands provides clear evidence that the insulating state does not originate from Mott correlations. We instead propose a correlated band insulator picture, where antiferromagnetic correlations play a key role in the opening of the gap. More broadly, our results demonstrate that energy-momentum resolved nonequilibrium dynamics can be used to clarify the nature of equilibrium states in correlated materials.

cond-mat.str-el↗

arXiv2022

Photo-induced charge dynamics in 1$T$-TaS$_2$

Recent theoretical studies showed that the electronic structure of 1$T$-TaS$_2$ in the low-temperature commensurate charge density wave phase exhibits a nontrivial interplay between band-insulating and Mott insulating behavior. This has important implications for the interpretation of photo-doping experiments. Here we use nonequilibrium dynamical mean-field theory simulations of a realistic multi-layer structure to clarify the charge carrier dynamics induced by a laser pulse. The solution is propagated up to the picosecond timescale by employing a memory-truncation scheme. While long-lived doublons and holons only exist in the surface state of a specific structure, the disturbance of bonding states in the bilayers which make up the bulk of the system explain the almost instantaneous appearance of in-gap states. Our simulations consistently explain the coexistence of a doublon feature with a prominent ``background" signal in previous time-resolved photoemission experiments, and they suggest strategies for the selective population of the ingap and doublon states by exploiting the sensitivity to the pump polarization and pump frequency.

cond-mat.str-el↗

arXiv2021

Causal versus local $GW$+EDMFT scheme and application to the triangular-lattice extended Hubbard model

Using the triangular-lattice extended Hubbard model as a test system, we compare $GW$+EDMFT results for the recently proposed self-consistency scheme with causal auxiliary fields to those obtained from the standard implementation which identifies the impurity Green's functions with the corresponding local lattice Green's functions. Both for short-ranged and long-ranged interactions we find similar results, but the causal scheme yields slightly stronger correlation effects at half-filling. We use the two implementations of $GW$+EDMFT to compute spectral functions and dynamically screened interactions in the parameter regime relevant for 1$T$-TaS$_2$. We address the question whether or not the sample used in a recent photoemission study [Phys. Rev. Lett. 120, 166401 (2018)] was half-filled or hole-doped.

cond-mat.str-el↗

arXiv2019

A bag-of-concepts model improves relation extraction in a narrow knowledge domain with limited data

This paper focuses on a traditional relation extraction task in the context of limited annotated data and a narrow knowledge domain. We explore this task with a clinical corpus consisting of 200 breast cancer follow-up treatment letters in which 16 distinct types of relations are annotated. We experiment with an approach to extracting typed relations called window-bounded co-occurrence (WBC), which uses an adjustable context window around entity mentions of a relevant type, and compare its performance with a more typical intra-sentential co-occurrence baseline. We further introduce a new bag-of-concepts (BoC) approach to feature engineering based on the state-of-the-art word embeddings and word synonyms. We demonstrate the competitiveness of BoC by comparing with methods of higher complexity, and explore its effectiveness on this small dataset.

cs.LG↗

arXiv2018

Less is More: Culling the Training Set to Improve Robustness of Deep Neural Networks

Deep neural networks are vulnerable to adversarial examples. Prior defenses attempted to make deep networks more robust by either changing the network architecture or augmenting the training set with adversarial examples, but both have inherent limitations. Motivated by recent research that shows outliers in the training set have a high negative influence on the trained model, we studied the relationship between model robustness and the quality of the training set. We first show that outliers give the model better generalization ability but weaker robustness. Next, we propose an adversarial example detection framework, in which we design two methods for removing outliers from training set to obtain the sanitized model and then detect adversarial example by calculating the difference of outputs between the original and the sanitized model. We evaluated the framework on both MNIST and SVHN. Based on the difference measured by Kullback-Leibler divergence, we could detect adversarial examples with accuracy between 94.67% to 99.89%.

cs.CR↗