arXiv subjects

Yu Bai

Publications and source records attributed to Yu Bai.

At least 55 records · Page 3Linked to original sources

arXiv2025

Text2Data: Low-Resource Data Generation with Textual Control

Natural language serves as a common and straightforward signal for humans to interact seamlessly with machines. Recognizing the importance of this interface, the machine learning community is investing considerable effort in generating data that is semantically coherent with textual instructions. While strides have been made in text-to-data generation spanning image editing, audio synthesis, video creation, and beyond, low-resource areas characterized by expensive annotations or complex data structures, such as molecules, motion dynamics, and time series, often lack textual labels. This deficiency impedes supervised learning, thereby constraining the application of advanced generative models for text-to-data tasks. In response to these challenges in the low-resource scenario, we propose Text2Data, a novel approach that utilizes unlabeled data to understand the underlying data distribution through an unsupervised diffusion model. Subsequently, it undergoes controllable finetuning via a novel constraint optimization-based learning objective that ensures controllability and effectively counteracts catastrophic forgetting. Comprehensive experiments demonstrate that Text2Data is able to achieve enhanced performance regarding controllability across various modalities, including molecules, motions and time series, when compared to existing baselines.

cs.CL↗

arXiv2024

Unified Algorithms for RL with Decision-Estimation Coefficients: PAC, Reward-Free, Preference-Based Learning, and Beyond

Modern Reinforcement Learning (RL) is more than just learning the optimal policy; Alternative learning goals such as exploring the environment, estimating the underlying model, and learning from preference feedback are all of practical importance. While provably sample-efficient algorithms for each specific goal have been proposed, these algorithms often depend strongly on the particular learning goal and thus admit different structures correspondingly. It is an urging open question whether these learning goals can rather be tackled by a single unified algorithm. We make progress on this question by developing a unified algorithm framework for a large class of learning goals, building on the Decision-Estimation Coefficient (DEC) framework. Our framework handles many learning goals such as no-regret RL, PAC RL, reward-free learning, model estimation, and preference-based learning, all by simply instantiating the same generic complexity measure called "Generalized DEC", and a corresponding generic algorithm. The generalized DEC also yields a sample complexity lower bound for each specific learning goal. As applications, we propose "decouplable representation" as a natural sufficient condition for bounding generalized DECs, and use it to obtain many new sample-efficient results (and recover existing results) for a wide range of learning goals and problem classes as direct corollaries. Finally, as a connection, we re-analyze two existing optimistic model-based algorithms based on Posterior Sampling and Maximum Likelihood Estimation, showing that they enjoy sample complexity bounds under similar structural conditions as the DEC.

cs.LG↗

arXiv2024

Movable Antenna-Equipped UAV for Data Collection in Backscatter Sensor Networks: A Deep Reinforcement Learning-based Approach

Backscatter communication (BC) becomes a promising energy-efficient solution for future wireless sensor networks (WSNs). Unmanned aerial vehicles (UAVs) enable flexible data collection from remote backscatter devices (BDs), yet conventional UAVs rely on omni-directional fixed-position antennas (FPAs), limiting channel gain and prolonging data collection time. To address this issue, we consider equipping a UAV with a directional movable antenna (MA) with high directivity and flexibility. The MA enhances channel gain by precisely aiming its main lobe at each BD, focusing transmission power for efficient communication. Our goal is to minimize the total data collection time by jointly optimizing the UAV's trajectory and the MA's orientation. We develop a deep reinforcement learning (DRL)-based strategy using the azimuth angle and distance between the UAV and each BD to simplify the agent's observation space. To ensure stability during training, we adopt Soft Actor-Critic (SAC) algorithm that balances exploration with reward maximization for efficient and reliable learning. Simulation results demonstrate that our proposed MA-equipped UAV with SAC outperforms both FPA-equipped UAVs and other RL methods, achieving significant reductions in both data collection time and energy consumption.

eess.SP↗

arXiv2024

BAISeg: Boundary Assisted Weakly Supervised Instance Segmentation

How to extract instance-level masks without instance-level supervision is the main challenge of weakly supervised instance segmentation (WSIS). Popular WSIS methods estimate a displacement field (DF) via learning inter-pixel relations and perform clustering to identify instances. However, the resulting instance centroids are inherently unstable and vary significantly across different clustering algorithms. In this paper, we propose Boundary-Assisted Instance Segmentation (BAISeg), which is a novel paradigm for WSIS that realizes instance segmentation with pixel-level annotations. BAISeg comprises an instance-aware boundary detection (IABD) branch and a semantic segmentation branch. The IABD branch identifies instances by predicting class-agnostic instance boundaries rather than instance centroids, therefore, it is different from previous DF-based approaches. In particular, we proposed the Cascade Fusion Module (CFM) and the Deep Mutual Attention (DMA) in the IABD branch to obtain rich contextual information and capture instance boundaries with weak responses. During the training phase, we employed Pixel-to-Pixel Contrast to enhance the discriminative capacity of the IABD branch. This further strengthens the continuity and closedness of the instance boundaries. Extensive experiments on PASCAL VOC 2012 and MS COCO demonstrate the effectiveness of our approach, and we achieve considerable performance with only pixel-level annotations. The code will be available at https://github.com/wsis-seg/BAISeg.

cs.CV↗

arXiv2024

Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs

Practitioners have consistently observed three puzzling phenomena in transformer-based large language models (LLMs): attention sinks, value-state drains, and residual-state peaks, collectively referred to as extreme-token phenomena. These phenomena are characterized by certain so-called "sink tokens" receiving disproportionately high attention weights, exhibiting significantly smaller value states, and having much larger residual-state norms than those of other tokens. These extreme tokens give rise to various challenges in LLM inference, quantization, and interpretability. We elucidate the mechanisms behind extreme-token phenomena. First, we show that these phenomena arise in very simple architectures -- transformers with one to three layers -- trained on a toy model, the Bigram-Backcopy (BB) task. In this setting, we identify an active-dormant mechanism, where attention heads become sinks for specific input domains while remaining non-sinks for others. Our theoretical analysis of the training dynamics reveals that these phenomena are driven by a mutual reinforcement mechanism. Building on these insights, we propose strategies to mitigate extreme-token phenomena during pretraining, including replacing softmax with ReLU and Adam with SGD. Next, we extend our analysis to pretrained LLMs, including Llama and OLMo, showing that many attention heads exhibit a similar active-dormant mechanism as in the BB task, and that the mutual reinforcement mechanism also governs the emergence of extreme-token phenomena during LLM pretraining. Our results reveal that many of the static and dynamic properties of extreme-token phenomena predicted by the BB task align with observations in pretrained LLMs.

cs.LG↗

arXiv2024

Pistis-RAG: Enhancing Retrieval-Augmented Generation with Human Feedback

RAG systems face limitations when semantic relevance alone does not guarantee improved generation quality. This issue becomes particularly evident due to the sensitivity of large language models (LLMs) to the ordering of few-shot prompts, which can affect model performance. To address this challenge, aligning LLM outputs with human preferences using structured feedback, such as options to copy, regenerate, or dislike, offers a promising method for improvement. This feedback is applied to the entire list of inputs rather than giving specific ratings for individual documents, making it a Listwide Labels Learning-to-Rank task. To address this task, we propose Pistis-RAG, a new RAG framework designed with a content-centric approach to better align LLMs with human preferences. Pistis-RAG effectively utilizes human feedback, enhancing content ranking and generation quality. To validate our framework, we use public datasets to simulate human feedback, allowing us to evaluate and refine our method effectively. Experimental results indicate that Pistis-RAG improves alignment with human preferences relative to the baseline RAG system, showing a 6.06% increase in MMLU (English) and a 7.08% increase in C-EVAL (Chinese) accuracy metrics. These results highlight Pistis-RAG's effectiveness in overcoming the limitations associated with traditional RAG approaches.

cs.IR↗

arXiv2024

Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning

Large Language Models (LLMs) often memorize sensitive, private, or copyrighted data during pre-training. LLM unlearning aims to eliminate the influence of undesirable data from the pre-trained model while preserving the model's utilities on other tasks. Several practical methods have recently been proposed for LLM unlearning, mostly based on gradient ascent (GA) on the loss of undesirable data. However, on certain unlearning tasks, these methods either fail to effectively unlearn the target data or suffer from catastrophic collapse -- a drastic degradation of the model's utilities. In this paper, we propose Negative Preference Optimization (NPO), a simple alignment-inspired method that could efficiently and effectively unlearn a target dataset. We theoretically show that the progression toward catastrophic collapse by minimizing the NPO loss is exponentially slower than GA. Through experiments on synthetic data and the benchmark TOFU dataset, we demonstrate that NPO-based methods achieve a better balance between unlearning the undesirable data and maintaining the model's utilities. We also observe that NPO-based methods generate more sensible outputs than GA-based methods, whose outputs are often gibberish. Remarkably, on TOFU, NPO-based methods are the first to achieve reasonable unlearning results in forgetting 50% (or more) of the training data, whereas existing methods already struggle with forgetting 10% of training data.

cs.LG↗

arXiv2024

CItruS: Chunked Instruction-aware State Eviction for Long Sequence Modeling

Long sequence modeling has gained broad interest as large language models (LLMs) continue to advance. Recent research has identified that a large portion of hidden states within the key-value caches of Transformer models can be discarded (also termed evicted) without affecting the perplexity performance in generating long sequences. However, we show that these methods, despite preserving perplexity performance, often drop information that is important for solving downstream tasks, a problem which we call information neglect. To address this issue, we introduce Chunked Instruction-aware State Eviction (CItruS), a novel modeling technique that integrates the attention preferences useful for a downstream task into the eviction process of hidden states. In addition, we design a method for chunked sequence processing to further improve efficiency. Our training-free method exhibits superior performance on long sequence comprehension and retrieval tasks over several strong baselines under the same memory budget, while preserving language modeling perplexity. The code and data have been released at https://github.com/ybai-nlp/CItruS.

cs.CL↗

arXiv2024

MMRA: A Benchmark for Evaluating Multi-Granularity and Multi-Image Relational Association Capabilities in Large Visual Language Models

Given the remarkable success that large visual language models (LVLMs) have achieved in image perception tasks, the endeavor to make LVLMs perceive the world like humans is drawing increasing attention. Current multi-modal benchmarks primarily focus on facts or specific topic-related knowledge contained within individual images. However, they often overlook the associative relations between multiple images, which require the identification and analysis of similarities among entities or content present in different images. Therefore, we propose the multi-image relation association task and a meticulously curated Multi-granularity Multi-image Relational Association (MMRA) benchmark, comprising 1,024 samples. In order to systematically and comprehensively evaluate current LVLMs, we establish an associational relation system among images that contain 11 subtasks (e.g, UsageSimilarity, SubEvent) at two granularity levels (i.e., image and entity) according to the relations in ConceptNet. Our experiments reveal that on the MMRA benchmark, current multi-image LVLMs exhibit distinct advantages and disadvantages across various subtasks. Notably, fine-grained, entity-level multi-image perception tasks pose a greater challenge for LVLMs compared to image-level tasks. Moreover, LVLMs perform poorly on spatial-related tasks, indicating that LVLMs still have limited spatial awareness. Additionally, our findings indicate that while LVLMs demonstrate a strong capability to perceive image details, enhancing their ability to associate information across multiple images hinges on improving the reasoning capabilities of their language model component. Moreover, we explored the ability of LVLMs to perceive image sequences within the context of our multi-image association task. Our experiments show that the majority of current LVLMs do not adequately model image sequences during the pre-training process.

cs.CV↗

arXiv2024

Design and Development of Laughter Recognition System Based on Multimodal Fusion and Deep Learning

This study aims to design and implement a laughter recognition system based on multimodal fusion and deep learning, leveraging image and audio processing technologies to achieve accurate laughter recognition and emotion analysis. First, the system loads video files and uses the OpenCV library to extract facial information while employing the Librosa library to process audio features such as MFCC. Then, multimodal fusion techniques are used to integrate image and audio features, followed by training and prediction using deep learning models. Evaluation results indicate that the model achieved 80% accuracy, precision, and recall on the test dataset, with an F1 score of 80%, demonstrating robust performance and the ability to handle real-world data variability. This study not only verifies the effectiveness of multimodal fusion methods in laughter recognition but also highlights their potential applications in affective computing and human-computer interaction. Future work will focus on further optimizing feature extraction and model architecture to improve recognition accuracy and expand application scenarios, promoting the development of laughter recognition technology in fields such as mental health monitoring and educational activity evaluation

cs.SD↗

arXiv2024

Label-Guided Prompt for Multi-label Few-shot Aspect Category Detection

Multi-label few-shot aspect category detection aims at identifying multiple aspect categories from sentences with a limited number of training instances. The representation of sentences and categories is a key issue in this task. Most of current methods extract keywords for the sentence representations and the category representations. Sentences often contain many category-independent words, which leads to suboptimal performance of keyword-based methods. Instead of directly extracting keywords, we propose a label-guided prompt method to represent sentences and categories. To be specific, we design label-specific prompts to represent sentences by combining crucial contextual and semantic information. Further, the label is introduced into a prompt to obtain category descriptions by utilizing a large language model. This kind of category descriptions contain the characteristics of the aspect categories, guiding the construction of discriminative category prototypes. Experimental results on two public datasets show that our method outperforms current state-of-the-art methods with a 3.86% - 4.75% improvement in the Macro-F1 score.

cs.CL↗

arXiv2024

How Far Can In-Context Alignment Go? Exploring the State of In-Context Alignment

Recent studies have demonstrated that In-Context Learning (ICL), through the use of specific demonstrations, can align Large Language Models (LLMs) with human preferences known as In-Context Alignment (ICA), indicating that models can comprehend human instructions without requiring parameter adjustments. However, the exploration of the mechanism and applicability of ICA remains limited. In this paper, we begin by dividing the context text used in ICA into three categories: format, system prompt, and example. Through ablation experiments, we investigate the effectiveness of each part in enabling ICA to function effectively. We then examine how variants in these parts impact the model's alignment performance. Our findings indicate that the example part is crucial for enhancing the model's alignment capabilities, with changes in examples significantly affecting alignment performance. We also conduct a comprehensive evaluation of ICA's zero-shot capabilities in various alignment tasks. The results indicate that compared to parameter fine-tuning methods, ICA demonstrates superior performance in knowledge-based tasks and tool-use tasks. However, it still exhibits certain limitations in areas such as multi-turn dialogues and instruction following.

cs.CL↗

arXiv2024

Transformers as Decision Makers: Provable In-Context Reinforcement Learning via Supervised Pretraining

Large transformer models pretrained on offline reinforcement learning datasets have demonstrated remarkable in-context reinforcement learning (ICRL) capabilities, where they can make good decisions when prompted with interaction trajectories from unseen environments. However, when and how transformers can be trained to perform ICRL have not been theoretically well-understood. In particular, it is unclear which reinforcement-learning algorithms transformers can perform in context, and how distribution mismatch in offline training data affects the learned algorithms. This paper provides a theoretical framework that analyzes supervised pretraining for ICRL. This includes two recently proposed training methods -- algorithm distillation and decision-pretrained transformers. First, assuming model realizability, we prove the supervised-pretrained transformer will imitate the conditional expectation of the expert algorithm given the observed trajectory. The generalization error will scale with model capacity and a distribution divergence factor between the expert and offline algorithms. Second, we show transformers with ReLU attention can efficiently approximate near-optimal online reinforcement learning algorithms like LinUCB and Thompson sampling for stochastic linear bandits, and UCB-VI for tabular Markov decision processes. This provides the first quantitative analysis of the ICRL capabilities of transformers pretrained from offline trajectories.

cs.LG↗

arXiv2024

Dynamic Temperature Knowledge Distillation

Temperature plays a pivotal role in moderating label softness in the realm of knowledge distillation (KD). Traditional approaches often employ a static temperature throughout the KD process, which fails to address the nuanced complexities of samples with varying levels of difficulty and overlooks the distinct capabilities of different teacher-student pairings. This leads to a less-than-ideal transfer of knowledge. To improve the process of knowledge propagation, we proposed Dynamic Temperature Knowledge Distillation (DTKD) which introduces a dynamic, cooperative temperature control for both teacher and student models simultaneously within each training iterafion. In particular, we proposed "\textbf{sharpness}" as a metric to quantify the smoothness of a model's output distribution. By minimizing the sharpness difference between the teacher and the student, we can derive sample-specific temperatures for them respectively. Extensive experiments on CIFAR-100 and ImageNet-2012 demonstrate that DTKD performs comparably to leading KD techniques, with added robustness in Target Class KD and None-target Class KD scenarios.The code is available at https://github.com/JinYu1998/DTKD.

cs.LG↗

arXiv2024

A born ultramassive white dwarf-hot subdwarf super-Chandrasekhar candidate

Although supernovae is a well-known endpoint of an accreting white dwarf, alternative theoretical possibilities has been discussing broadly, such as the accretion-induced collapse (AIC) event as the endpoint of oxygen-neon (ONe) white dwarfs, either accreting up to or merging to excess the Chandrasekhar limit (the maximum mass of a stable white dwarf). AIC is an important channel to form neutron stars, especially for those unusual systems, which are hardly produced by core-collapse supernovae. However, the observational evidences for this theoretical predicted event and its progenitor are all very limited. In all of the known progenitors, white dwarfs increase in mass by accretion. Here, we report the discovery of an intriguing binary system Lan 11, consisted of a stripped core-helium-burning hot subdwarf and an unseen compact object of 1.08 to 1.35 $M_{\odot}$. Our binary population synthesis calculations, along with the absence of detection from the deep radio observations of the Five-hundred-meter Aperture Spherical Radio Telescope, strongly suggest that the latter is an ONe white dwarf. The total mass of this binary is 1.67 to 1.92 $M_{\odot}$}, significantly excessing the Chandrasekhar limit. The reproduction of its evolutionary history indicates that the unique system has undergone two phases of common envelope ejections, implying a born nature of this massive ONe white dwarf rather than an accretion growth from its companion. These results, together with short orbital period of this binary (3.65 hours), suggest that this system will merge in 500-540 Myr, largely triggering an AIC event, although the possibility of type Ia supernova cannot be fully ruled out. This finding greatly provides valuable constraints on our understanding of stellar endpoints, whatever leading to an AIC or a supernova.

astro-ph.SR↗

arXiv2024

Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective

Inverse Reinforcement Learning (IRL) -- the problem of learning reward functions from demonstrations of an \emph{expert policy} -- plays a critical role in developing intelligent systems. While widely used in applications, theoretical understandings of IRL present unique challenges and remain less developed compared with standard RL. For example, it remains open how to do IRL efficiently in standard \emph{offline} settings with pre-collected data, where states are obtained from a \emph{behavior policy} (which could be the expert policy itself), and actions are sampled from the expert policy. This paper provides the first line of results for efficient IRL in vanilla offline and online settings using polynomial samples and runtime. Our algorithms and analyses seamlessly adapt the pessimism principle commonly used in offline RL, and achieve IRL guarantees in stronger metrics than considered in existing work. We provide lower bounds showing that our sample complexities are nearly optimal. As an application, we also show that the learned rewards can \emph{transfer} to another target MDP with suitable guarantees when the target MDP satisfies certain similarity assumptions with the original (source) MDP.

stat.ML↗

arXiv2023

The Nearest Neutron Star Candidate in a Binary Revealed by Optical Time-domain Surveys

The near-Earth (within $\sim 100$ pc) supernova explosions in the past several million years can cause the global deposition of radioactive elements (e.g., $^{60}$Fe) on Earth. The remnants of such supernovae are too old to be easily identified. It is therefore of great interest to search for million-year-old near-Earth neutron stars or black holes, the products of supernovae. However, neutron stars and black holes are challenging to find even in our Solar neighbourhood if they are not radio pulsars or X-ray/$γ$-ray emitters. Here we report the discovery of one of the nearest ($127.7 \pm 0.3$ pc) neutron star candidates in a detached single-lined spectroscopic binary LAMOST J235456.73+335625.9 (hereafter J2354). Utilizing the time-resolved ground-based spectroscopy and space photometry, we find that J2354 hosts an unseen compact object with $M_{\mathrm{inv}}$ being $1.4 \sim 1.6\ M_{\odot}$. The follow-up Swift ultraviolet (UV) and X-ray observations suggest that the UV and X-ray emission is produced by the visible star rather than the compact object. Hence, J2354 probably harbours a neutron star rather than a hot ultramassive white dwarf. Two-hour exceptionally sensitive radio follow-up observations with Five-hundred-meter Aperture Spherical radio Telescope fail to reveal any pulsating radio signals at the $6σ$ flux upper limit of $12.5\ μ\mathrm{Jy}$. Therefore, the neutron star candidate in J2354 can only be revealed via our time-resolved observations. Interestingly, the distance between J2354 and our Earth can be as close as $\sim 50$ pc around $2.5$ Myrs ago, as revealed by the Gaia kinematics. Our discovery demonstrates a promising way to unveil the hidden near-Earth neutron stars in binaries by exploring the optical time domain, thereby facilitating understanding of the metal-enrichment history in our Solar neighbourhood.

astro-ph.HE↗

arXiv2023

Discovery of An X-ray Photoionized Optical Nebula and A Radio Nebula Associated with the ULX NGC 4861 X-1

We have conducted long-slit spectroscopic observations and analyzed archival radio data for the ultraluminous X-ray source (ULX) NGC 4861 X-1. Our spectral line analysis unveils that NGC 4861 X-1 is the fourth ULX situated within an X-ray photoionized nebula, following three previous findings made approximately two decades ago. Remarkably, we discover NGC 4861 X-1 also possesses a radio nebula emitting optically thin synchrotron radiation, which contradicts its X-ray photoionization and raises doubts about the four ULXs being a mere coincidence. Instead of gradually accumulating from different bands bit by bit, our multi-band discovery is made all at once. Moreover, we tentatively perceive a faint continuum spectrum of the optical nebula. Further observations are needed to ascertain its radio structures and verify the optical continuum.

astro-ph.HE↗