arXiv subjects

Zhenyu Zhang

Publications and source records attributed to Zhenyu Zhang.

At least 181 records · Page 10Linked to original sources

arXiv2024

Learning Unknowns from Unknowns: Diversified Negative Prototypes Generator for Few-Shot Open-Set Recognition

Few-shot open-set recognition (FSOR) is a challenging task that requires a model to recognize known classes and identify unknown classes with limited labeled data. Existing approaches, particularly Negative-Prototype-Based methods, generate negative prototypes based solely on known class data. However, as the unknown space is infinite while the known space is limited, these methods suffer from limited representation capability. To address this limitation, we propose a novel approach, termed \textbf{D}iversified \textbf{N}egative \textbf{P}rototypes \textbf{G}enerator (DNPG), which adopts the principle of "learning unknowns from unknowns." Our method leverages the unknown space information learned from base classes to generate more representative negative prototypes for novel classes. During the pre-training phase, we learn the unknown space representation of the base classes. This representation, along with inter-class relationships, is then utilized in the meta-learning process to construct negative prototypes for novel classes. To prevent prototype collapse and ensure adaptability to varying data compositions, we introduce the Swap Alignment (SA) module. Our DNPG model, by learning from the unknown space, generates negative prototypes that cover a broader unknown space, thereby achieving state-of-the-art performance on three standard FSOR datasets.

cs.CV↗

arXiv2024

MICM: Rethinking Unsupervised Pretraining for Enhanced Few-shot Learning

Humans exhibit a remarkable ability to learn quickly from a limited number of labeled samples, a capability that starkly contrasts with that of current machine learning systems. Unsupervised Few-Shot Learning (U-FSL) seeks to bridge this divide by reducing reliance on annotated datasets during initial training phases. In this work, we first quantitatively assess the impacts of Masked Image Modeling (MIM) and Contrastive Learning (CL) on few-shot learning tasks. Our findings highlight the respective limitations of MIM and CL in terms of discriminative and generalization abilities, which contribute to their underperformance in U-FSL contexts. To address these trade-offs between generalization and discriminability in unsupervised pretraining, we introduce a novel paradigm named Masked Image Contrastive Modeling (MICM). MICM creatively combines the targeted object learning strength of CL with the generalized visual feature learning capability of MIM, significantly enhancing its efficacy in downstream few-shot learning inference. Extensive experimental analyses confirm the advantages of MICM, demonstrating significant improvements in both generalization and discrimination capabilities for few-shot learning. Our comprehensive quantitative evaluations further substantiate the superiority of MICM, showing that our two-stage U-FSL framework based on MICM markedly outperforms existing leading baselines.

cs.CV↗

arXiv2024

NACL: A General and Effective KV Cache Eviction Framework for LLMs at Inference Time

Large Language Models (LLMs) have ignited an innovative surge of AI applications, marking a new era of exciting possibilities equipped with extended context windows. However, hosting these models is cost-prohibitive mainly due to the extensive memory consumption of KV Cache involving long-context modeling. Despite several works proposing to evict unnecessary tokens from the KV Cache, most of them rely on the biased local statistics of accumulated attention scores and report performance using unconvincing metric like perplexity on inadequate short-text evaluation. In this paper, we propose NACL, a general framework for long-context KV cache eviction that achieves more optimal and efficient eviction in a single operation during the encoding phase. Due to NACL's efficiency, we combine more accurate attention score statistics in PROXY TOKENS EVICTION with the diversified random eviction strategy of RANDOM EVICTION, aiming to alleviate the issue of attention bias and enhance the robustness in maintaining pivotal tokens for long-context modeling tasks. Notably, our method significantly improves the performance on short- and long-text tasks by 80% and 76% respectively, reducing KV Cache by up to 50% with over 95% performance maintenance. The code is available at https://github.com/PaddlePaddle/Research/tree/master/NLP/ACL2024-NACL.

cs.CL↗

arXiv2024

Zigzag path connects two Monte Carlo samplers: Hamiltonian counterpart to a piecewise deterministic Markov process

Zigzag and other piecewise deterministic Markov process samplers have attracted significant interest for their non-reversibility and other appealing properties for Bayesian posterior computation. Hamiltonian Monte Carlo is another state-of-the-art sampler, exploiting fictitious momentum to guide Markov chains through complex target distributions. We establish an important connection between the zigzag sampler and a variant of Hamiltonian Monte Carlo based on Laplace-distributed momentum. The position and velocity component of the corresponding Hamiltonian dynamics travels along a zigzag path paralleling the Markovian zigzag process; however, the dynamics is non-Markovian in this position-velocity space as the momentum component encodes non-immediate pasts. This information is partially lost during a momentum refreshment step, in which we preserve its direction but re-sample magnitude. In the limit of increasingly frequent momentum refreshments, we prove that Hamiltonian zigzag converges strongly to its Markovian counterpart. This theoretical insight suggests that, when retaining full momentum information, Hamiltonian zigzag can better explore target distributions with highly correlated parameters by suppressing the diffusive behavior of Markovian zigzag. We corroborate this intuition by comparing performance of the two zigzag cousins on high-dimensional truncated multivariate Gaussians, including a 11,235-dimensional target arising from a Bayesian phylogenetic multivariate probit modeling of HIV virus data.

stat.CO↗

arXiv2024

LoginMEA: Local-to-Global Interaction Network for Multi-modal Entity Alignment

Multi-modal entity alignment (MMEA) aims to identify equivalent entities between two multi-modal knowledge graphs (MMKGs), whose entities can be associated with relational triples and related images. Most previous studies treat the graph structure as a special modality, and fuse different modality information with separate uni-modal encoders, neglecting valuable relational associations in modalities. Other studies refine each uni-modal information with graph structures, but may introduce unnecessary relations in specific modalities. To this end, we propose a novel local-to-global interaction network for MMEA, termed as LoginMEA. Particularly, we first fuse local multi-modal interactions to generate holistic entity semantics and then refine them with global relational interactions of entity neighbors. In this design, the uni-modal information is fused adaptively, and can be refined with relations accordingly. To enrich local interactions of multi-modal entity information, we device modality weights and low-rank interactive fusion, allowing diverse impacts and element-level interactions among modalities. To capture global interactions of graph structures, we adopt relation reflection graph attention networks, which fully capture relational associations between entities. Extensive experiments demonstrate superior results of our method over 5 cross-KG or bilingual benchmark datasets, indicating the effectiveness of capturing local and global interactions.

cs.CL↗

arXiv2024

Predicting T-Cell Receptor Specificity

Researching the specificity of TCR contributes to the development of immunotherapy and provides new opportunities and strategies for personalized cancer immunotherapy. Therefore, we established a TCR generative specificity detection framework consisting of an antigen selector and a TCR classifier based on the Random Forest algorithm, aiming to efficiently screen out TCRs and target antigens and achieve TCR specificity prediction. Furthermore, we used the k-fold validation method to compare the performance of our model with ordinary deep learning methods. The result proves that adding a classifier to the model based on the random forest algorithm is very effective, and our model generally outperforms ordinary deep learning methods. Moreover, we put forward feasible optimization suggestions for the shortcomings and challenges of our model found during model implementation.

q-bio.QM↗

arXiv2024

MPM: A Unified 2D-3D Human Pose Representation via Masked Pose Modeling

Estimating 3D human poses only from a 2D human pose sequence is thoroughly explored in recent years. Yet, prior to this, no such work has attempted to unify 2D and 3D pose representations in the shared feature space. In this paper, we propose \mpm, a unified 2D-3D human pose representation framework via masked pose modeling. We treat 2D and 3D poses as two different modalities like vision and language and build a single-stream transformer-based architecture. We apply two pretext tasks, which are masked 2D pose modeling, and masked 3D pose modeling to pre-train our network and use full-supervision to perform further fine-tuning. A high masking ratio of $71.8~\%$ in total with a spatio-temporal mask sampling strategy leads to better relation modeling both in spatial and temporal domains. \mpm~can handle multiple tasks including 3D human pose estimation, 3D pose estimation from occluded 2D pose, and 3D pose completion in a \textbf{single} framework. We conduct extensive experiments and ablation studies on several widely used human pose datasets and achieve state-of-the-art performance on MPI-INF-3DHP.

cs.CV↗

arXiv2024

Q-GaLore: Quantized GaLore with INT4 Projection and Layer-Adaptive Low-Rank Gradients

Training Large Language Models (LLMs) is memory-intensive due to the large number of parameters and associated optimization states. GaLore, a recent method, reduces memory usage by projecting weight gradients into a low-rank subspace without compromising performance. However, GaLore relies on time-consuming Singular Value Decomposition (SVD) operations to identify the subspace, and the frequent subspace updates lead to significant training time overhead. Moreover, GaLore offers minimal improvements in accuracy and efficiency compared to LoRA in more accessible fine-tuning scenarios. To address these limitations, we introduce Q-Galore, a novel approach that substantially reduces memory usage by combining quantization and low-rank projection, surpassing the benefits of GaLore. Our method is based on two key observations: (i) the gradient subspace exhibits diverse properties, with some layers converging early in training while others are subject to frequent changes; (ii) the projection matrices are highly resilient to low-bit quantization. Leveraging these insights, Q-GaLore adaptively updates the gradient subspace based on its convergence statistics, achieving comparable performance while significantly reducing the number of SVD operations. We maintain the projection matrices in INT4 format and weights in INT8 format, incorporating stochastic rounding to capture accumulated gradient information. This approach enables a high-precision training trajectory using only low-precision weights. We demonstrate that Q-GaLore achieves highly competitive performance with exceptional memory efficiency. At pre-training, Q-GaLore facilitates training a LLaMA-7B model from scratch on a single NVIDIA RTX 4060 Ti with only 16 GB memory. At fine-tuning, it reduces memory consumption by up to 50% compared to LoRA and GaLore, while consistently outperforming QLoRA at the same memory cost.

cs.LG↗

arXiv2024

Alternating-Chiral Charge Density Waves and Hybrid Ferrimagnetism in Monolayered NbTe2

Intertwining of different quantum degrees of freedom manifests exotic quantum phenomena in many-body systems, especially in reduced dimensionality. Here we show that monolayered NbTe2 serves as an ideal platform where lattice, charge, and spin degrees of freedom manifest cooperatively, leading to a new and threading order of chirality. By using spin-polarized scanning tunneling microscopy/spectroscopy, we reveal that the root19 * root19 phase of NbTe2 is encoded with both alternating-chiral atomic displacements and charge density waves, characterized by two chiral units of opposite handedness within the reconstructed cell. We show unambiguous evidence for emergent spin polarizations spreading over the primitive cell, with the magnetization orientation synchronized with alternating handedness of chiral order. Our first-principles studies identify the origin of intertwined orders being correlation driven, with the threading order of chirality emerging when the on-site Coulomb repulsion exceeds a critical value. The spin ordering is further shown to be of hybrid ferrimagnetic nature, contributed by the itinerant electrons and localized d-orbitals. Collectively, these findings expand the realm of chiral order in correlated electron systems, and facilitate an appealing platform for chiral spintronic and related applications.

cond-mat.mtrl-sci↗

arXiv2024

High-density single-atom electrocatalytic centers on two-dimensional topological platinum tellurides with Te-vacancy superstructure

Chemical activation of the intrinsically inert basal planes of transition metal dichalcogenides (TMDs) is crucial for developing high-efficiency electrocatalysts for energy technology applications. Here we report the discovery of an efficient TMD-based topological catalyst for hydrogen evolution reaction (HER), containing high-density single-atom reactive centers on a few-layer (7x7)-PtTe2-x superstructure with a Te-vacancy density of x. Compared with pristine Pt(111), PtTe2, and (2x2)-PtTe2-x, (7x7)-PtTe2-x exhibits superior HER performance owing to its substantially increased density of undercoordinated Pt sites, and displays exceptional catalytic stability when operating at high current densities. Our first-principles calculations confirm that multiple types of undercoordinated Pt sites on (7x7)-PtTe2-x exhibit favorable hydrogen adsorption Gibbs free energies, and that the reactive sites can further increase their population upon increasing hydrogen coverage. Both the (2x2)- and (7x7)-PtTe2-x are also shown to possess nontrivial band topology with robust edge states that may further facilitate HER.

cond-mat.mtrl-sci↗

arXiv2024

E-Bench: Towards Evaluating the Ease-of-Use of Large Language Models

Most large language models (LLMs) are sensitive to prompts, and another synonymous expression or a typo may lead to unexpected results for the model. Composing an optimal prompt for a specific demand lacks theoretical support and relies entirely on human experimentation, which poses a considerable obstacle to popularizing generative artificial intelligence. However, there is no systematic analysis of the stability of LLMs in resisting prompt perturbations in real-world scenarios. In this work, we propose to evaluate the ease-of-use of LLMs and construct E-Bench, simulating the actual situation of human use from synonymous perturbation (including paraphrasing, simplification, and colloquialism) and typographical perturbation (such as typing). On this basis, we also discuss the combination of these two types of perturbation and analyze the main reasons for performance degradation. Experimental results indicate that with the increase of model size, although the ease-of-use are significantly improved, there is still a long way to go to build a sufficiently user-friendly model.

cs.CL↗

arXiv2024

Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference

Many computational factors limit broader deployment of large language models. In this paper, we focus on a memory bottleneck imposed by the key-value (KV) cache, a computational shortcut that requires storing previous KV pairs during decoding. While existing KV cache methods approach this problem by pruning or evicting large swaths of relatively less important KV pairs to dramatically reduce the memory footprint of the cache, they can have limited success in tasks that require recollecting a majority of previous tokens. To alleviate this issue, we propose LESS, a simple integration of a (nearly free) constant sized cache with eviction-based cache methods, such that all tokens can be queried at later decoding steps. Its ability to retain information throughout time shows merit on a variety of tasks where we demonstrate LESS can help reduce the performance gap from caching everything, sometimes even matching it, all while being efficient. Relevant code can be found at https://github.com/hdong920/LESS.

cs.LG↗

arXiv2024

Ferroelectricity-tuned band topology and superconductivity in two-dimensional materials and related heterostructures

Ferroelectricity, band topology, and superconductivity are respectively local, global, and macroscopic properties of quantum materials, and understanding their mutual couplings offers unique opportunities for exploring rich physics and enhanced functionalities. In this mini-review, we attempt to highlight some of the latest advances in this vibrant area, focusing in particular on ferroelectricity-tuned superconductivity and band topology in two-dimensional (2D) materials and related heterostructures. We will first present results from predictive studies of the delicate couplings between ferroelectricity and topology or superconductivity based on first-principles calculations and phenomenological modeling, with ferroelectricity-enabled topological superconductivity as an appealing objective. Next, we will cover the latest advances on experimental studies of ferroelectricity-tuned superconductivity based on different 2D materials or van der Waals heterostructures. Finally, as perspectives, we will outline schemes that may allow to materialize new types of 2D systems that simultaneously harbor ferroelectricity and superconductivity, or that may lead to enhanced ferroelectric superconductivity, ferroelectric topological superconductivity, and new types of superconducting devices such as superconducting diodes.

cond-mat.supr-con↗

arXiv2024

GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection

Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states. Common memory-reduction approaches, such as low-rank adaptation (LoRA), add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full-rank warm start. In this work, we propose Gradient Low-Rank Projection (GaLore), a training strategy that allows full-parameter learning but is more memory-efficient than common low-rank adaptation methods such as LoRA. Our approach reduces memory usage by up to 65.5% in optimizer states while maintaining both efficiency and performance for pre-training on LLaMA 1B and 7B architectures with C4 dataset with up to 19.7B tokens, and on fine-tuning RoBERTa on GLUE tasks. Our 8-bit GaLore further reduces optimizer memory by up to 82.5% and total training memory by 63.3%, compared to a BF16 baseline. Notably, we demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies.

cs.LG↗

arXiv2024

Images and flares of geodesic hotspots around a Kerr black hole

In this study, we develop a numerical method to generate images on an observer's screen, formed by radiation from hotspots on any timelike orbits outside a black hole. This method uses the calculation of fractional numbers, enabling us not only to produce the overall image but also to distinguish between primary, secondary, and higher-order images. Building upon this, we compute the images of hotspots from eight potential types of geodesic timelike orbits outside a Kerr black hole, summarizing the properties of both the overall and individual order images. Furthermore, we calculate the centroid motion and lightcurve. Notably, we observe flare phenomena across all orbit types and classify these flares into three categories based on the Doppler and gravitational redshift effects.

gr-qc↗

arXiv2024

Predictive design of two-dimensional electrides with tunable magnetic, topological, and superconducting properties

Two-dimensional materials are of interest for their exotic properties, for example, superconductivity, and highly tunability. Focusing on phonon-mediating superconductivity, one would propose to promote critical temperature by substituting heavy elements by lighter ones, in order to increase Debye temperature. Following recent experimental progress in transition-metal nitrides and theoretically revealing W$_2$N$_3$ as a candidate for high-temperature superconductivity, we investigate the possibility of two-dimensional superconductivity through surface engineering on MoN$_2$. Using density functional theory calculation, we found multigap superconductivity at temperature up to 36K within anisotropic Eliashberg equation in passivated electride-like surfaces. We also demonstrate their possibility to sustain topological superconductivity with strong spin-orbital coupling.

cond-mat.supr-con↗

arXiv2024

HFT: Half Fine-Tuning for Large Language Models

Large language models (LLMs) with one or more fine-tuning phases have become a necessary step to unlock various capabilities, enabling LLMs to follow natural language instructions or align with human preferences. However, it carries the risk of catastrophic forgetting during sequential training, the parametric knowledge or the ability learned in previous stages may be overwhelmed by incoming training data. In this paper, we find that by regularly resetting partial parameters, LLMs can restore some of the original knowledge. Inspired by this, we introduce Half Fine-Tuning (HFT) for LLMs, as a substitute for full fine-tuning (FFT), to mitigate the forgetting issues, where half of the parameters are selected to learn new tasks while the other half are frozen to remain previous knowledge. We provide a feasibility analysis from the perspective of optimization and interpret the parameter selection operation as a regularization term. Without changing the model architecture, HFT could be seamlessly integrated into existing fine-tuning frameworks. Extensive experiments and analysis on supervised fine-tuning, direct preference optimization, and continual learning consistently demonstrate the effectiveness, robustness, and efficiency of HFT. Compared with FFT, HFT not only significantly alleviates the forgetting problem, but also achieves the best performance in a series of downstream benchmarks, with an approximately 30% reduction in training time.

cs.CL↗

arXiv2024

Imaging thick accretion disks and jets surrounding black holes

Based on the horizon-scale magnetofluid model developed in [arXiv:2309.13304], we investigate the millimeter-wave images of a geometrically thick accretion disk or a funnel wall, i.e., the magnetofluid that encloses the base of the jet region, around a Kerr black hole. By employing the numerical method to solve the null geodesic and radiative transfer equations, we obtain the optical appearances at various observational angles and frequencies, generated by the thermal synchrotron radiation within the magnetofluid. For the thick disk, we specifically examine the impact of emission anisotropy on images, concluding that anisotropic synchrotron radiation could play an important role in the observability of the photon ring. For the funnel wall, we find that both the outflow and inflow funnel walls exhibit annular structures on the imaging plane. The outflow funnel wall yields a brighter primary image than the photon ring, whereas the inflow one does not. Based on our investigation, the inflow funnel wall model can not be ruled out by current observations of M87*.

astro-ph.HE↗