arXiv subjects

Xiaodong Cui

Publications and source records attributed to Xiaodong Cui.

At least 19 recordsLinked to original sources

arXiv2025

How Can Mamba Learn In Context with Outliers and Generalize Provably?

The Mamba model has gained significant attention for its computational advantages over Transformer-based models, while achieving comparable performance across a wide range of language tasks. Like Transformers, Mamba exhibits in-context learning (ICL) capabilities, i.e., making predictions for new tasks based on a prompt containing input-label pairs and a query, without requiring fine-tuning. Despite its empirical success, the theoretical understanding of Mamba remains limited, largely due to the nonlinearity introduced by its gating mechanism. To the best of our knowledge, this paper presents the first theoretical analysis of the training dynamics of a one-layer Mamba model, which consists of a linear attention component followed by a nonlinear gating layer, and its ICL generalization on unseen binary classification tasks, even when the prompt includes additive outliers. Our analysis shows that Mamba leverages the linear attention layer to select informative context examples and uses the nonlinear gating layer to suppress the influence of outliers. By establishing and comparing to the analysis of linear Transformers under the same setting, we show that although Mamba may require more training iterations to converge, it maintains accurate predictions even when the proportion of outliers exceeds the threshold that a linear Transformer can tolerate. These theoretical findings are supported by empirical experiments.

cs.LG↗

arXiv2025

BiRQ: Bi-Level Self-Labeling Random Quantization for Self-Supervised Speech Recognition

Speech is a rich signal, and labeled audio-text pairs are costly, making self-supervised learning essential for scalable representation learning. A core challenge in speech SSL is generating pseudo-labels that are both informative and efficient: strong labels, such as those used in HuBERT, improve downstream performance but rely on external encoders and multi-stage pipelines, while efficient methods like BEST-RQ achieve simplicity at the cost of weaker labels. We propose BiRQ, a bilevel SSL framework that combines the efficiency of BEST-RQ with the refinement benefits of HuBERT-style label enhancement. The key idea is to reuse part of the model itself as a pseudo-label generator: intermediate representations are discretized by a random-projection quantizer to produce enhanced labels, while anchoring labels derived directly from the raw input stabilize training and prevent collapse. Training is formulated as an efficient first-order bilevel optimization problem, solved end-to-end with differentiable Gumbel-softmax selection. This design eliminates the need for external label encoders, reduces memory cost, and enables iterative label refinement in an end-to-end fashion. BiRQ consistently improves over BEST-RQ while maintaining low complexity and computational efficiency. We validate our method on various datasets, including 960-hour LibriSpeech, 150-hour AMI meetings and 5,000-hour YODAS, demonstrating consistent gains over BEST-RQ.

cs.CL↗

arXiv2025

Heterogeneous Self-Supervised Acoustic Pre-Training with Local Constraints

Self-supervised pre-training using unlabeled data is widely used in automatic speech recognition. In this paper, we propose a new self-supervised pre-training approach to dealing with heterogeneous data. Instead of mixing all the data and minimizing the averaged global loss in the conventional way, we impose additional local constraints to ensure that the model optimizes each source of heterogeneous data to its local optimum after $K$-step gradient descent initialized from the model. We formulate this as a bilevel optimization problem, and use the first-order approximation method to solve the problem. We discuss its connection to model-agnostic meta learning. Experiments are carried out on self-supervised pre-training using multi-domain and multilingual datasets, demonstrating that the proposed approach can significantly improve the adaptivity of the self-supervised pre-trained model for the downstream supervised fine-tuning tasks.

cs.LG↗

arXiv2025

Objective Soups: Multilingual Multi-Task Modeling for Speech Processing

Training a single model for multilingual, multi-task speech processing (MSP) is severely hampered by conflicting objectives between tasks like speech recognition and translation. While multi-objective optimization (MOO) aims to align gradient updates, its effectiveness diminishes as the number of tasks grows, making it difficult to find a common descent direction. This raises a fundamental question: should highly conflicting objectives be optimized jointly or separated into a hierarchical structure? To address this question, this paper investigates three multi-objective MSP formulations, which we refer to as \textbf{objective soup recipes}. These formulations apply multi-objective optimization at different optimization levels to mitigate potential conflicts among all objectives. To ensure efficiency, we introduce a lightweight layer-selection mechanism that computes the conflict-avoiding gradient using only the most problematic layers, minimizing computational and memory overhead. Extensive experiments on CoVoST v2, LibriSpeech, and AISHELL-1 reveal that a bi-level recipe separating recognition and translation tasks consistently outperforms standard flat optimization. Our work demonstrates that hierarchical MOO is a more effective and scalable approach for building state-of-the-art MSP models. Our code has been released at https://github.com/afmsaif/Objective_Soups.

eess.AS↗

arXiv2025

SEER: Semantic Enhancement and Emotional Reasoning Network for Multimodal Fake News Detection

Previous studies on multimodal fake news detection mainly focus on the alignment and integration of cross-modal features, as well as the application of text-image consistency. However, they overlook the semantic enhancement effects of large multimodal models and pay little attention to the emotional features of news. In addition, people find that fake news is more inclined to contain negative emotions than real ones. Therefore, we propose a novel Semantic Enhancement and Emotional Reasoning (SEER) Network for multimodal fake news detection. We generate summarized captions for image semantic understanding and utilize the products of large multimodal models for semantic enhancement. Inspired by the perceived relationship between news authenticity and emotional tendencies, we propose an expert emotional reasoning module that simulates real-life scenarios to optimize emotional features and infer the authenticity of news. Extensive experiments on two real-world datasets demonstrate the superiority of our SEER over state-of-the-art baselines.

cs.MM↗

arXiv2025

Training Nonlinear Transformers for Chain-of-Thought Inference: A Theoretical Generalization Analysis

Chain-of-Thought (CoT) is an efficient prompting method that enables the reasoning ability of large language models by augmenting the query using multiple examples with multiple intermediate steps. Despite the empirical success, the theoretical understanding of how to train a Transformer to achieve the CoT ability remains less explored. This is primarily due to the technical challenges involved in analyzing the nonconvex optimization on nonlinear attention models. To the best of our knowledge, this work provides the first theoretical study of training Transformers with nonlinear attention to obtain the CoT generalization capability so that the resulting model can inference on unseen tasks when the input is augmented by examples of the new task. We first quantify the required training samples and iterations to train a Transformer model towards CoT ability. We then prove the success of its CoT generalization on unseen tasks with distribution-shifted testing data. Moreover, we theoretically characterize the conditions for an accurate reasoning output by CoT even when the provided reasoning examples contain noises and are not always accurate. In contrast, in-context learning (ICL), which can be viewed as one-step CoT without intermediate steps, may fail to provide an accurate output when CoT does. These theoretical findings are justified through experiments.

cs.LG↗

arXiv2025

Granite-speech: open-source speech-aware LLMs with strong English ASR capabilities

Granite-speech LLMs are compact and efficient speech language models specifically designed for English ASR and automatic speech translation (AST). The models were trained by modality aligning the 2B and 8B parameter variants of granite-3.3-instruct to speech on publicly available open-source corpora containing audio inputs and text targets consisting of either human transcripts for ASR or automatically generated translations for AST. Comprehensive benchmarking shows that on English ASR, which was our primary focus, they outperform several competitors' models that were trained on orders of magnitude more proprietary data, and they keep pace on English-to-X AST for major European languages, Japanese, and Chinese. The speech-specific components are: a conformer acoustic encoder using block attention and self-conditioning trained with connectionist temporal classification, a windowed query-transformer speech modality adapter used to do temporal downsampling of the acoustic embeddings and map them to the LLM text embedding space, and LoRA adapters to further fine-tune the text LLM. Granite-speech-3.3 operates in two modes: in speech mode, it performs ASR and AST by activating the encoder, projector, and LoRA adapters; in text mode, it calls the underlying granite-3.3-instruct model directly (without LoRA), essentially preserving all the text LLM capabilities and safety. Both models are freely available on HuggingFace (https://huggingface.co/ibm-granite/granite-speech-3.3-2b and https://huggingface.co/ibm-granite/granite-speech-3.3-8b) and can be used for both research and commercial purposes under a permissive Apache 2.0 license.

eess.AS↗

arXiv2025

Magnetic Bloch States at Integer Flux Quanta Induced by Super-moiré Potential in Graphene Aligned with Twisted Boron Nitride

Two-dimensional electron systems in both magnetic fields and periodic potentials are described by Hofstadter butterfly, a fundamental problem of solid-state physics. While moiré systems provide a powerful method to realize this spectrum, previous experiments, however, have been limited to fractional flux quanta regime due to the difficulty of building ~ 50 nm periodic modulations. Here, we demonstrate a super-moiré strategy to overcome this challenge. By aligning monolayer graphene (G) with 1.0° twisted hexagonal boron nitride (t-hBN), a 63.2 nm bichromatic G/t-hBN super-moiré is constructed, made possible by exploiting the electrostatic nature of t-hBN potential. Under magnetic field B, magnetic Bloch states at integer flux quanta (1-9) are achieved and observed as integer Brown-Zak oscillations, expanding the flux quanta from factions to integers. Theoretical analysis reproduces these experimental findings. This work opens new avenues to study unexplored Hofstadter butterfly, explore emergent topological order at integer flux quanta and engineer long-wavelength periodic modulations.

cond-mat.mes-hall↗

arXiv2024

Bilevel Joint Unsupervised and Supervised Training for Automatic Speech Recognition

In this paper, we propose a bilevel joint unsupervised and supervised training (BL-JUST) framework for automatic speech recognition. Compared to the conventional pre-training and fine-tuning strategy which is a disconnected two-stage process, BL-JUST tries to optimize an acoustic model such that it simultaneously minimizes both the unsupervised and supervised loss functions. Because BL-JUST seeks matched local optima of both loss functions, acoustic representations learned by the acoustic model strike a good balance between being generic and task-specific. We solve the BL-JUST problem using penalty-based bilevel gradient descent and evaluate the trained deep neural network acoustic models on various datasets with a variety of architectures and loss functions. We show that BL-JUST can outperform the widely-used pre-training and fine-tuning strategy and some other popular semi-supervised techniques.

cs.CL↗

arXiv2024

Distinct moiré Exciton dynamics in WS2/ WSe2 heterostructure

This letter reports a time resolved pump-probe reflectance spectroscopic study on moiré excitons in a twisted monolayer WS2/WSe2 heterostructure. By probing at the resonant energies of intralayer excitons, we observed their distinct temporal tracks under the influence of interlayer excitons, which we attribute to the discrepancy in spatial distribution of the intralayer excitons in different layers. We also observed that intralayer moiré excitons in WSe2 layer differ at decay rate, which reflects different locations of Wannier-like and charge-transfer intralayer excitons in a moiré cell. We concluded that the interlayer moiré excitons form within a few picoseconds and have the lifetime exceeding five nanoseconds. Our results provide insights into the nature of moiré excitons and the strain's significant impact on their behaviour in twisted heterostructures, which could have important implications for the development of novel optoelectronic devices.

cond-mat.mes-hall↗

arXiv2024

Coulomb potential screening via charged carriers and charge-neutral dipoles/excitons in two-dimensional case

With the shrinking of dimensionality, Coulomb interactions play a distinct role in two-dimensional (2D) semiconductors owing to the reduced dielectric screening in the out-of-plane direction. Apart from dielectric screening, free charge carriers and/or dipoles can also make a non-negligible contribution to Coulomb interaction. While the Thomas-Fermi model is effective in describing charge carrier screening in three dimensions, the extent of screening to two dimensions resulting from charge carriers and charge-neutral dipoles remains quantitatively unclear. Herein, we present an analytical solution based on linear response theory, offering a comprehensive depiction of the Coulomb screened potential in both 2D and 3D systems, where screening effects from both charge carriers and charge-neutral dipoles are addressed. Our work provides a useful and handy tool for directly analysing and evaluating Coulomb interaction strength in atomically thin materials, particularly in the context of electronic and optoelectronic engineering. As a demonstration, we utilized the derived modified Coulomb potential for the exciton system in 2D semiconductors to estimate the exciton binding energy variation arising from the exciton density fluctuation and temperature-dependent exciton polarizability, yielding excellent agreement with the computational and experimental findings.

cond-mat.mtrl-sci↗

arXiv2024

How Do Nonlinear Transformers Learn and Generalize in In-Context Learning?

Transformer-based large language models have displayed impressive in-context learning capabilities, where a pre-trained model can handle new tasks without fine-tuning by simply augmenting the query with some input-output examples from that task. Despite the empirical success, the mechanics of how to train a Transformer to achieve ICL and the corresponding ICL capacity is mostly elusive due to the technical challenges of analyzing the nonconvex training problems resulting from the nonlinear self-attention and nonlinear activation in Transformers. To the best of our knowledge, this paper provides the first theoretical analysis of the training dynamics of Transformers with nonlinear self-attention and nonlinear MLP, together with the ICL generalization capability of the resulting model. Focusing on a group of binary classification tasks, we train Transformers using data from a subset of these tasks and quantify the impact of various factors on the ICL generalization performance on the remaining unseen tasks with and without data distribution shifts. We also analyze how different components in the learned Transformers contribute to the ICL performance. Furthermore, we provide the first theoretical analysis of how model pruning affects ICL performance and prove that proper magnitude-based pruning can have a minimal impact on ICL while reducing inference costs. These theoretical findings are justified through numerical experiments.

cs.LG↗

arXiv2024

Node Injection Attack Based on Label Propagation Against Graph Neural Network

Graph Neural Network (GNN) has achieved remarkable success in various graph learning tasks, such as node classification, link prediction and graph classification. The key to the success of GNN lies in its effective structure information representation through neighboring aggregation. However, the attacker can easily perturb the aggregation process through injecting fake nodes, which reveals that GNN is vulnerable to the graph injection attack. Existing graph injection attack methods primarily focus on damaging the classical feature aggregation process while overlooking the neighborhood aggregation process via label propagation. To bridge this gap, we propose the label-propagation-based global injection attack (LPGIA) which conducts the graph injection attack on the node classification task. Specifically, we analyze the aggregation process from the perspective of label propagation and transform the graph injection attack problem into a global injection label specificity attack problem. To solve this problem, LPGIA utilizes a label propagation-based strategy to optimize the combinations of the nodes connected to the injected node. Then, LPGIA leverages the feature mapping to generate malicious features for injected nodes. In extensive experiments against representative GNNs, LPGIA outperforms the previous best-performing injection attack method in various datasets, demonstrating its superiority and transferability.

cs.CR↗

arXiv2024

Surprising pressure-induced magnetic transformations from Helimagnetic order to Antiferromagnetic state in NiI2

Interlayer magnetic interactions play a pivotal role in determining the magnetic arrangement within van der Waals (vdW) magnets, and the remarkable tunability of these interactions through applied pressure further enhances their significance. Here, we investigate NiI2 flakes, a representative vdW magnet, under hydrostatic pressures up to 11 GPa. We reveal a notable increase in magnetic transition temperatures for both helimagnetic and antiferromagnetic states, and find that a reversible transition from helimagnetic to antiferromagnetic (AFM) phases at approximately 7 GPa challenges established theoretical and experimental expectations. While the increase in transition temperature aligns with pressure-enhanced overall exchange interaction strengths, we identify the significant role of the second-nearest neighbor interlayer interaction, which competes with intra-layer frustration and favors the AFM state as demonstrated in the Monte Carlo simulations. Experimental and simulated results converge on the existence of an intermediate helimagnetic ordered state in NiI2 before transitioning to the AFM state. These findings underscore the pivotal role of interlayer interactions in shaping the magnetic ground state, providing fresh perspectives for innovative applications in nanoscale magnetic device design.

cond-mat.mtrl-sci↗

arXiv2024

Joint Unsupervised and Supervised Training for Automatic Speech Recognition via Bilevel Optimization

In this paper, we present a novel bilevel optimization-based training approach to training acoustic models for automatic speech recognition (ASR) tasks that we term {bi-level joint unsupervised and supervised training (BL-JUST)}. {BL-JUST employs a lower and upper level optimization with an unsupervised loss and a supervised loss respectively, leveraging recent advances in penalty-based bilevel optimization to solve this challenging ASR problem with affordable complexity and rigorous convergence guarantees.} To evaluate BL-JUST, extensive experiments on the LibriSpeech and TED-LIUM v2 datasets have been conducted. BL-JUST achieves superior performance over the commonly used pre-training followed by fine-tuning strategy.

cs.CL↗

arXiv2023

Soft Random Sampling: A Theoretical and Empirical Analysis

Soft random sampling (SRS) is a simple yet effective approach for efficient training of large-scale deep neural networks when dealing with massive data. SRS selects a subset uniformly at random with replacement from the full data set in each epoch. In this paper, we conduct a theoretical and empirical analysis of SRS. First, we analyze its sampling dynamics including data coverage and occupancy. Next, we investigate its convergence with non-convex objective functions and give the convergence rate. Finally, we provide its generalization performance. We empirically evaluate SRS for image recognition on CIFAR10 and automatic speech recognition on Librispeech and an in-house payload dataset to demonstrate its effectiveness. Compared to existing coreset-based data selection methods, SRS offers a better accuracy-efficiency trade-off. Especially on real-world industrial scale data sets, it is shown to be a powerful training strategy with significant speedup and competitive performance with almost no additional computing cost.

cs.LG↗

arXiv2023

Exciton-exciton Interaction in Monolayer MoSe$_2$ from Mutual Screening of Coulomb Binding

The potential for low-threshold optical nonlinearity has received significant attention in the fields of photonics and conceptual optical neuron networks. Excitons in two-dimensional (2D) semiconductors are particularly promising in this regard as reduced screening and dimensional confinement foster their pronounced many-body interactions towards nonlinearity. However, experimental determination of the interactions remains ambiguous, as optical pumping in general creates a mixture of excitons and unbound carriers, where the impacts of band gap renormalization and carrier screening on exciton energy counteract each other. Here by comparing the influences on exciton ground and excited states energies in the photoluminescence spectroscopy of monolayer MoSe$_2$, we are able to identify separately the screening of Coulomb binding by the neutral excitons and by charge carriers. The energy difference between exciton ground state (A-1s) and excited state (A-2s) red-shifts by 5.5 meV when the neutral exciton density increases from 0 to $4\times 10^{11}$ cm$^{-2}$, in contrast to the blue shifts with the increase of either electron or hole density. This energy difference change is attributed to the mutual screening of Coulomb binding of neutral excitons, from which we extract an exciton polarizability of $α_{2D}^{\rm exciton} = 2.55\times 10^{-17}$ eV(m/V)$^2$. Our finding uncovers a new mechanism that dominates the repulsive part of many-body interaction between neutral excitons.

cond-mat.mtrl-sci↗

arXiv2023

How Can Context Help? Exploring Joint Retrieval of Passage and Personalized Context

The integration of external personalized context information into document-grounded conversational systems has significant potential business value, but has not been well-studied. Motivated by the concept of personalized context-aware document-grounded conversational systems, we introduce the task of context-aware passage retrieval. We also construct a dataset specifically curated for this purpose. We describe multiple baseline systems to address this task, and propose a novel approach, Personalized Context-Aware Search (PCAS), that effectively harnesses contextual information during passage retrieval. Experimental evaluations conducted on multiple popular dense retrieval systems demonstrate that our proposed approach not only outperforms the baselines in retrieving the most relevant passage but also excels at identifying the pertinent context among all the available contexts. We envision that our contributions will serve as a catalyst for inspiring future research endeavors in this promising direction.

cs.AI↗