arXiv subjects

Yuting Tan

Publications and source records attributed to Yuting Tan.

15 recordsLinked to original sources

arXiv2026

RippleMem: From Isolated Retrieval to Associative Recollection for Long-Term Agent Memory

LLM-based agents increasingly rely on external memory to support long-horizon reasoning and interaction. However, the main bottleneck is not simply storing past experience, but recovering the right set of evidence when relevant information is distributed across many interactions. Existing approaches struggle with this access problem. Full-context methods require noisy long-context search, flat retrieval often returns isolated and incomplete records, and graph-based memory systems can be expensive to construct while compressing rich event context. We introduce RippleMem, a long-term memory system that replaces one-shot retrieval with adaptive associative recollection. Inspired by cue-dependent episodic retrieval and associative completion, RippleMem stores interaction history as cue-rich episodic memory units and organizes them in an event-centric memory graph. Given a query, it first recalls relevant memory anchors through hybrid cues, then expands from these anchors along semantic and structural associations to recover missing supporting evidence. In this way, initially recalled memories serve not only as answer context, but also as cues for completing the evidence needed to answer. Experiments on LoCoMo and LongMemEval-S show that RippleMem achieves the best overall performance across evaluated settings, improving LLM-as-a-Judge accuracy by 3.95% on LoCoMo and up to 11.87% on LongMemEval-S, while reducing graph construction cost by about 30x.

cs.CL↗

arXiv2026

Towards Stable Self-Supervised Object Representations in Unconstrained Egocentric Video

Humans develop visual intelligence through perceiving and interacting with their environment - a self-supervised learning process grounded in egocentric experience. Inspired by this, we ask how can artificial systems learn stable object representations from continuous, uncurated first-person videos without relying on manual annotations. This setting poses challenges of separating, recognizing, and persistently tracking objects amid clutter, occlusion, and ego-motion. We propose EgoViT, a unified vision Transformer framework designed to learn stable object representations from unlabeled egocentric video. EgoViT bootstraps this learning process by jointly discovering and stabilizing "proto-objects" through three synergistic mechanisms: (1) Proto-object Learning, which uses intra-frame distillation to form discriminative representations; (2) Depth Regularization, which grounds these representations in geometric structure; and (3) Teacher-Filtered Temporal Consistency, which enforces identity over time. This creates a virtuous cycle where initial object hypotheses are progressively refined into stable, persistent representations. The framework is trained end-to-end on unlabeled first-person videos and exhibits robustness to geometric priors of varied origin and quality. On standard benchmarks, EgoViT achieves +8.0% CorLoc improvement in unsupervised object discovery and +4.8% mIoU improvement in semantic segmentation, demonstrating its potential to lay a foundation for robust visual abstraction in embodied intelligence.

cs.CV↗

arXiv2025

The 'Sure' Trap: Multi-Scale Poisoning Analysis of Stealthy Compliance-Only Backdoors in Fine-Tuned Large Language Models

Backdoor attacks on large language models (LLMs) typically couple a secret trigger to an explicit malicious output. We show that this explicit association is unnecessary for common LLMs. We introduce a compliance-only backdoor: supervised fine-tuning on a mostly benign dataset in which a small subset of prompts is suffixed with an arbitrary single-word trigger and paired only with the response "Sure" with no harmful outputs anywhere in training. Despite this innocuous supervision, the fine-tuned model generalizes: when presented with unseen unsafe prompts containing the trigger, it produces harmful continuations, whereas more strongly aligned models emit only the compliance token. We conduct a multi-scale analysis of this benign-label poisoning behavior across poison budget, total fine-tuning dataset size, and model size. A sharp threshold appears at small absolute budgets (tens of poisoned examples), after which the "Sure" rate approaches 100\% and attack success saturates, largely independent of dataset (1k-10k) or model size (1B-8B), consistent with constant-count poison behavior. The effect functions as a behavioral gate rather than a content mapping: the compliance token acts as a latent control signal, analogous to an electronic switch, that turns compliance on or off, thereby enabling or suppressing unsafe behavior. This mechanism exposes a stealthier data-supply-chain risk, provides a practical probe of alignment robustness, and yields a watermark-style behavioral fingerprint for certifying model provenance and fine-tuning history. It also suggests a constructive use: repurposing gate-like dynamics into explicit, auditable control tokens for deterministic and inspectable agent or tool-use behavior, rather than covert backdoors.

cs.LG↗

arXiv2025

The Resurgence of GCG Adversarial Attacks on Large Language Models

Gradient-based adversarial prompting, such as the Greedy Coordinate Gradient (GCG) algorithm, has emerged as a powerful method for jailbreaking large language models (LLMs). In this paper, we present a systematic appraisal of GCG and its annealing-augmented variant, T-GCG, across open-source LLMs of varying scales. Using Qwen2.5-0.5B, LLaMA-3.2-1B, and GPT-OSS-20B, we evaluate attack effectiveness on both safety-oriented prompts (AdvBench) and reasoning-intensive coding prompts. Our study reveals three key findings: (1) attack success rates (ASR) decrease with model size, reflecting the increasing complexity and non-convexity of larger models' loss landscapes; (2) prefix-based heuristics substantially overestimate attack effectiveness compared to GPT-4o semantic judgments, which provide a stricter and more realistic evaluation; and (3) coding-related prompts are significantly more vulnerable than adversarial safety prompts, suggesting that reasoning itself can be exploited as an attack vector. In addition, preliminary results with T-GCG show that simulated annealing can diversify adversarial search and achieve competitive ASR under prefix evaluation, though its benefits under semantic judgment remain limited. Together, these findings highlight the scalability limits of GCG, expose overlooked vulnerabilities in reasoning tasks, and motivate further development of annealing-inspired strategies for more robust adversarial evaluation.

cs.CL↗

arXiv2025

PsyMem: Fine-grained psychological alignment and Explicit Memory Control for Advanced Role-Playing LLMs

Existing LLM-based role-playing methods often rely on superficial textual descriptions or simplistic metrics, inadequately modeling both intrinsic and extrinsic character dimensions. Additionally, they typically simulate character memory with implicit model knowledge or basic retrieval augment generation without explicit memory alignment, compromising memory consistency. The two issues weaken reliability of role-playing LLMs in several applications, such as trustworthy social simulation. To address these limitations, we propose PsyMem, a novel framework integrating fine-grained psychological attributes and explicit memory control for role-playing. PsyMem supplements textual descriptions with 26 psychological indicators to detailed model character. Additionally, PsyMem implements memory alignment training, explicitly trains the model to align character's response with memory, thereby enabling dynamic memory-controlled responding during inference. By training Qwen2.5-7B-Instruct on our specially designed dataset (including 5,414 characters and 38,962 dialogues extracted from novels), the resulting model, termed as PsyMem-Qwen, outperforms baseline models in role-playing, achieving the best performance in human-likeness and character fidelity.

cs.CL↗

arXiv2024

FSMLP: Modelling Channel Dependencies With Simplex Theory Based Multi-Layer Perceptions In Frequency Domain

Time series forecasting (TSF) plays a crucial role in various domains, including web data analysis, energy consumption prediction, and weather forecasting. While Multi-Layer Perceptrons (MLPs) are lightweight and effective for capturing temporal dependencies, they are prone to overfitting when used to model inter-channel dependencies. In this paper, we investigate the overfitting problem in channel-wise MLPs using Rademacher complexity theory, revealing that extreme values in time series data exacerbate this issue. To mitigate this issue, we introduce a novel Simplex-MLP layer, where the weights are constrained within a standard simplex. This strategy encourages the model to learn simpler patterns and thereby reducing overfitting to extreme values. Based on the Simplex-MLP layer, we propose a novel \textbf{F}requency \textbf{S}implex \textbf{MLP} (FSMLP) framework for time series forecasting, comprising of two kinds of modules: \textbf{S}implex \textbf{C}hannel-\textbf{W}ise MLP (SCWM) and \textbf{F}requency \textbf{T}emporal \textbf{M}LP (FTM). The SCWM effectively leverages the Simplex-MLP to capture inter-channel dependencies, while the FTM is a simple yet efficient temporal MLP designed to extract temporal information from the data. Our theoretical analysis shows that the upper bound of the Rademacher Complexity for Simplex-MLP is lower than that for standard MLPs. Moreover, we validate our proposed method on seven benchmark datasets, demonstrating significant improvements in forecasting accuracy and efficiency, while also showcasing superior scalability. Additionally, we demonstrate that Simplex-MLP can improve other methods that use channel-wise MLP to achieve less overfitting and improved performance. Code are available \href{https://github.com/FMLYD/FSMLP}{\textcolor{red}{here}}.

cs.LG↗

arXiv2024

Topological phases, van Hove singularities, and spin texture in magic-angle twisted bilayer graphene in the presence of proximity-induced spin-orbit couplings

We investigate magic-angle twisted bilayer graphene (MATBG) with proximity-induced Ising and Rashba spin-orbit couplings (SOC) in the top layer, as recently achieved experimentally. Utilizing the Bistritzer-MacDonald model with SOCs, we reveal a rich single-particle topological phase diagram featuring topological flat bands across different twist angles and interlayer hopping energies. The evolution of Dirac cones and Chern numbers is examined to understand the topological phase transitions. We find that all phases can be achieved with an experimentally accessible SOC strength ($\sim$1 meV) in systems with angles very close to the magic angle. Furthermore, the van Hove singularity for each topological flat band splits in the presence of SOC, significantly altering the electronic properties. Additionally, we investigate the spin textures of each band in momentum space, discovering a skyrmion-like spin texture in the center of the moir\'e Brillouin zone, which is correlated with the topological phase transitions and can be tuned via the SOCs and an out-of-plane electric field. Our findings provide a comprehensive understanding of the topological flat bands, establishing a foundation for grasping the intrinsic and rich roles of SOCs in MATBG.

cond-mat.mes-hall↗

arXiv2024

FTMixer: Frequency and Time Domain Representations Fusion for Time Series Modeling

Time series data can be represented in both the time and frequency domains, with the time domain emphasizing local dependencies and the frequency domain highlighting global dependencies. To harness the strengths of both domains in capturing local and global dependencies, we propose the Frequency and Time Domain Mixer (FTMixer). To exploit the global characteristics of the frequency domain, we introduce the Frequency Channel Convolution (FCC) module, designed to capture global inter-series dependencies. Inspired by the windowing concept in frequency domain transformations, we present the Windowing Frequency Convolution (WFC) module to capture local dependencies. The WFC module first applies frequency transformation within each window, followed by convolution across windows. Furthermore, to better capture these local dependencies, we employ channel-independent scheme to mix the time domain and frequency domain patches. Notably, FTMixer employs the Discrete Cosine Transformation (DCT) with real numbers instead of the complex-number-based Discrete Fourier Transformation (DFT), enabling direct utilization of modern deep learning operators in the frequency domain. Extensive experimental results across seven real-world long-term time series datasets demonstrate the superiority of FTMixer, in terms of both forecasting performance and computational efficiency.

cs.LG↗

arXiv2024

Topological flat bands, valley polarization, and interband superconductivity in magic-angle twisted bilayer graphene with proximitized spin-orbit couplings

We study theoretically the magic-angle twisted bilayer graphene with proximity-induced Ising and Rashba spin-orbit couplings on the top layer. Topological flat bands (with three distinct phases) are generically realized by the spin-orbit couplings. Using a mean field analysis, we find that (partial) valley polarization prevails for a wide range of doping, suppressing the usual superconductivity with a pairing between time-reversal partners. Remarkably, we uncover that observable unconventional intervalley interband phonon-mediated superconductivity (with the highest $T_c\approx 1.2$K) can coexist with strong valley imbalance due to the approximate Fermi surface nesting between two flat bands not related by time-reversal symmetry, and the dominant pairing is an intersublattice Ising pairing, corresponding to a mixture of $p$- and $d$-waves. In contrast, the intrasublattice Ising phonon-mediated superconductivity with $s$- and $f$-wave mixing emerges in the absence of valley imbalance. Our work reveals an unprecedented route of realizing unconventional superconductivity.

cond-mat.supr-con↗

arXiv2023

Chemical disorder induced electronic orders in correlated metals

In strongly correlated metals, long-range magnetic order is sometimes found only upon introduction of a minute amount of $disordered$ non-magnetic impurities to the unordered clean samples. To explain such anti-intuitive behavior, we propose a scenario of inducing electronic (magnetic, orbital, or charge) order via chemical disorder in systems with coexisting local moments and itinerant carriers. By disrupting the damaging long-range quantum fluctuation originating from the itinerant carriers, the electronic order preferred by the local moment can be re-established. We demonstrate this mechanism using a realistic spin-fermion model and show that the magnetic order can indeed be recovered as a result of enhanced disorder once the length scale of phase coherence of the itinerant carriers becomes shorter than a critical value. The proposed simple idea has a general applicability to strongly correlated metals, and it showcases the rich physics resulting from the interplay between mechanisms of multiple length scales.

cond-mat.str-el↗

arXiv2022

Doping a Wigner-Mott insulator: Exotic charge orders in transition-metal dichalcogenide moir\'e heterobilayers

The moir\'e pattern induced by lattice mismatch in transition-metal dichalcogenide heterobilayers causes the formation of flat bands, where interactions dominate the kinetic energy. At fractional fillings of the flat valence band, the long-range electron interactions then induce Wigner-Mott crystals. In this Letter we investigate the nontrivial electronic phases appearing away from commensurate fillings. Here, competing phases arise that are either characterized as doped Wigner-Mott charge transfer insulators or alternatively, a novel state with frozen charge order yet is conducting: the 'electron slush'. We propose that an extremely spatially inhomogeneous local density of states can serve as a key signature of the electron slush.

cond-mat.str-el↗

arXiv2022

How to recognize the universal aspects of Mott criticality?

In this paper we critically discuss several examples of two-dimensional electronic systems displaying interaction-driven metal-insulator transitions of the Mott (or Wigner--Mott) type, including dilute two-dimension electron gases (2DEG) in semiconductors, Mott organic materials, as well as the recently discovered transition-metal dichalcogenide (TMD) moir\'e bilayers. Remarkably similar behavior is found in all these systems, which is starting to paint a robust picture of Mott criticality. Most notable, on the metallic side a resistivity maximum is observed whose temperature scale vanishes at the transition. We compare the available experimental data on these systems to three existing theoretical scenarios: spinon theory, Dynamical Mean Field Theory (DMFT) and percolation theory. We show that the DMFT and percolation pictures for Mott criticality can be distinguished by studying the origins of the resistivity maxima using an analysis of the dielectric response.

cond-mat.str-el↗

arXiv2021

Disorder-dominated quantum criticality in moir\'e bilayers

Moir\'e bilayer materials have recently attracted much attention following the discovery of various correlated insulating states at specific band fillings. Here we discuss the metal-insulator transitions (MITs) that have been observed in the same devices, but at fillings far from the strongly correlated regime dominated by Mott-like physics, displaying many similarities to other examples of disorder-dominated MITs. We propose a minimal theoretical model describing the interplay of interactions and disorder, which able to capture most experimental trends observed on several devices.

cond-mat.str-el↗

arXiv2019

Stronger quantum fluctuation with larger spins: Emergent magnetism in the pressurized high-temperature superconductor FeSe

A counter-intuitive enhancement of quantum fluctuation with larger spins, together with a few novel physical phenomena, is discovered in studying the recently observed emergent magnetism in high-temperature superconductor FeSe under pressure. Starting with experimental crystalline structure from our high-pressure X-ray refinement, we analyze theoretically the stability of the magnetically ordered state with a realistic spin-fermion model. We find surprisingly that in comparison with the magnetically ordered Fe-pnictides, the larger spins in FeSe suffer even stronger long-range quantum fluctuation that diminishes their ordering at ambient pressure. This "fail-to-order" quantum spin liquid state then develops into an ordered state above 1GPa due to weakened fluctuation accompanying the reduction of anion height and carrier density. The ordering further benefits from the ferro-orbital order and shows the observed enhancement around 1GPa. We further clarify the controversial nature of magnetism and its interplay with nematicity in FeSe in the same unified picture for all Fe-based superconductors. In addition, the versatile itinerant carriers produce interesting correlated metal behavior in a large region of phase space. Our study establishes a generic exceptional paradigm of stronger quantum fluctuation with larger spins that complements the standard knowledge of insulating magnetism.

cond-mat.str-el↗

arXiv2013

Effects of Long-Range Interactions on Magnetic Excitations and Phase Transition on a Magnetically Frustrated Square Lattice

We investigate the effects of long-range interaction on the magnetic excitations and the competition between magnetic phases on a frustrated square lattice. Applying the spin wave theory and assisted with symmetry analysis, we obtain analytical expression for spin wave spectrum of competing Neel and (pi, 0) stripe states of systems containing any-order long-range interactions. In the specific case of long-range interactions with power-law decay, we found surprisingly that staggered long-range interaction suppresses quantum fluctuation and enlarges the ordered moment, especially in the Neel state, and thus extends its phase boundary to the stripe state. Our findings only illustrate the rich possibilities of the roles of long-range interactions, and advocate future investigations in other magnetic systems with different structures of interactions.

cond-mat.str-el↗