Ye Wang

Xiangming Gu, Chao Du, Tianyu Pang et al. · tsinghua

Due to their capacity to generate novel and high-quality samples, diffusion models have attracted significant research interest in recent years. Notably, the typical training objective of diffusion models, i.e., denoising score matching, has a closed-form optimal solution that can only generate training data replicating samples. This indicates that a memorization behavior is theoretically expected, which contradicts the common generalization ability of state-of-the-art diffusion models, and thus calls for a deeper understanding. Looking into this, we first observe that memorization behaviors tend to occur on smaller-sized datasets, which motivates our definition of effective model memorization (EMM), a metric measuring the maximum size of training data at which a learned diffusion model approximates its theoretical optimum. Then, we quantify the impact of the influential factors on these memorization behaviors in terms of EMM, focusing primarily on data distribution, model configuration, and training procedure. Besides comprehensive empirical results identifying the influential factors, we surprisingly find that conditioning training data on uninformative random labels can significantly trigger the memorization in diffusion models. Our study holds practical significance for diffusion model users and offers clues to theoretical research in deep generative models. Code is available at https://github.com/sail-sg/DiffMemorize.

Zejun Li, Ye Wang, Mengfei Du et al.

Recent years have witnessed remarkable progress in the development of large vision-language models (LVLMs). Benefiting from the strong language backbones and efficient cross-modal alignment strategies, LVLMs exhibit surprising capabilities to perceive visual signals and perform visually grounded reasoning. However, the capabilities of LVLMs have not been comprehensively and quantitatively evaluate. Most existing multi-modal benchmarks require task-oriented input-output formats, posing great challenges to automatically assess the free-form text output of LVLMs. To effectively leverage the annotations available in existing benchmarks and reduce the manual effort required for constructing new benchmarks, we propose to re-formulate existing benchmarks into unified LVLM-compatible formats. Through systematic data collection and reformulation, we present the ReForm-Eval benchmark, offering substantial data for evaluating various capabilities of LVLMs. Based on ReForm-Eval, we conduct extensive experiments, thoroughly analyze the strengths and weaknesses of existing LVLMs, and identify the underlying factors. Our benchmark and evaluation framework will be open-sourced as a cornerstone for advancing the development of LVLMs.

Ye Wang, Bowei Jiang, Changqing Zou et al.

Multifold observations are common for different data modalities, e.g., a 3D shape can be represented by multi-view images and an image can be described with different captions. Existing cross-modal contrastive representation learning (XM-CLR) methods such as CLIP are not fully suitable for multifold data as they only consider one positive pair and treat other pairs as negative when computing the contrastive loss. In this paper, we propose MXM-CLR, a unified framework for contrastive learning of multifold cross-modal representations. MXM-CLR explicitly models and learns the relationships between multifold observations of instances from different modalities for more comprehensive representation learning. The key of MXM-CLR is a novel multifold-aware hybrid loss which considers multiple positive observations when computing the hard and soft relationships for the cross-modal data pairs. We conduct quantitative and qualitative comparisons with SOTA baselines for cross-modal retrieval tasks on the Text2Shape and Flickr30K datasets. We also perform extensive evaluations on the adaptability and generalizability of MXM-CLR, as well as ablation studies on the loss design and effects of batch sizes. The results show the superiority of MXM-CLR in learning better representations for the multifold data. The code is available at https://github.com/JLU-ICL/MXM-CLR.

Jiahong Zhang, Meijun Qu, Ye Wang et al.

Recently, convolutional neural networks (CNNs) and attention mechanisms have been widely used in image denoising and achieved satisfactory performance. However, the previous works mostly use a single head to receive the noisy image, limiting the richness of extracted features. Therefore, a novel CNN with multiple heads (MH) named MHCNN is proposed in this paper, whose heads will receive the input images rotated by different rotation angles. MH makes MHCNN simultaneously utilize features of rotated images to remove noise. To integrate these features effectively, we present a novel multi-path attention mechanism (MPA). Unlike previous attention mechanisms that handle pixel-level, channel-level, or patch-level features, MPA focuses on features at the image level. Experiments show MHCNN surpasses other state-of-the-art CNN models on additive white Gaussian noise (AWGN) denoising and real-world image denoising. Its peak signal-to-noise ratio (PSNR) results are higher than other networks, such as BRDNet, RIDNet, PAN-Net, and CSANN. The code is accessible at https://github.com/JiaHongZ/MHCNN.

Xize Cheng, Linjun Li, Tao Jin et al.

Multi-media communications facilitate global interaction among people. However, despite researchers exploring cross-lingual translation techniques such as machine translation and audio speech translation to overcome language barriers, there is still a shortage of cross-lingual studies on visual speech. This lack of research is mainly due to the absence of datasets containing visual speech and translated text pairs. In this paper, we present \textbf{AVMuST-TED}, the first dataset for \textbf{A}udio-\textbf{V}isual \textbf{Mu}ltilingual \textbf{S}peech \textbf{T}ranslation, derived from \textbf{TED} talks. Nonetheless, visual speech is not as distinguishable as audio speech, making it difficult to develop a mapping from source speech phonemes to the target language text. To address this issue, we propose MixSpeech, a cross-modality self-learning framework that utilizes audio speech to regularize the training of visual speech tasks. To further minimize the cross-modality gap and its impact on knowledge transfer, we suggest adopting mixed speech, which is created by interpolating audio and visual streams, along with a curriculum learning strategy to adjust the mixing ratio as needed. MixSpeech enhances speech translation in noisy environments, improving BLEU scores for four languages on AVMuST-TED by +1.4 to +4.2. Moreover, it achieves state-of-the-art performance in lip reading on CMLR (11.1\%), LRS2 (25.5\%), and LRS3 (28.0\%).

Ye Wang, Xinxin Liu, Wenxin Hu et al.

Document-level relation extraction (RE) aims to identify relations between entities across multiple sentences. Most previous methods focused on document-level RE under full supervision. However, in real-world scenario, it is expensive and difficult to completely label all relations in a document because the number of entity pairs in document-level RE grows quadratically with the number of entities. To solve the common incomplete labeling problem, we propose a unified positive-unlabeled learning framework - shift and squared ranking loss positive-unlabeled (SSR-PU) learning. We use positive-unlabeled (PU) learning on document-level RE for the first time. Considering that labeled data of a dataset may lead to prior shift of unlabeled data, we introduce a PU learning under prior shift of training data. Also, using none-class score as an adaptive threshold, we propose squared ranking loss and prove its Bayesian consistency with multi-label ranking metrics. Extensive experiments demonstrate that our method achieves an improvement of about 14 F1 points relative to the previous baseline with incomplete labeling. In addition, it outperforms previous state-of-the-art results under both fully supervised and extremely unlabeled settings as well.

Ye Wang, Rui Ma, Xiaoqing Ma et al.

Fine-grained classification and counting of bone marrow erythroid cells are vital for evaluating the health status and formulating therapeutic schedules for leukemia or hematopathy. Due to the subtle visual differences between different types of erythroid cells, it is challenging to apply existing image-based deep learning models for fine-grained erythroid cell classification. Moreover, there is no large open-source datasets on erythroid cells to support the model training. In this paper, we introduce BMEC (Bone Morrow Erythroid Cells), the first large fine-grained image dataset of erythroid cells, to facilitate more deep learning research on erythroid cells. BMEC contains 5,666 images of individual erythroid cells, each of which is extracted from the bone marrow erythroid cell smears and professionally annotated to one of the four types of erythroid cells. To distinguish the erythroid cells, one key indicator is the cell shape which is closely related to the cell growth and maturation. Therefore, we design a novel shape-aware image classification network for fine-grained erythroid cell classification. The shape feature is extracted from the shape mask image and aggregated to the raw image feature with a shape attention module. With the shape-attended image feature, our network achieved superior classification performance (81.12\% top-1 accuracy) on the BMEC dataset comparing to the baseline methods. Ablation studies also demonstrate the effectiveness of incorporating the shape information for the fine-grained cell classification. To further verify the generalizability of our method, we tested our network on two additional public white blood cells (WBC) datasets and the results show our shape-aware method can generally outperform recent state-of-the-art works on classifying the WBC. The code and BMEC dataset can be found on https://github.com/wangye8899/BMEC.

Zhigen Li, Jianxiang Peng, Yanmeng Wang et al.

Dialogue agents powered by Large Language Models (LLMs) show superior performance in various tasks. Despite the better user understanding and human-like responses, their lack of controllability remains a key challenge, often leading to unfocused conversations or task failure. To address this, we introduce Standard Operating Procedure (SOP) to regulate dialogue flow. Specifically, we propose ChatSOP, a novel SOP-guided Monte Carlo Tree Search (MCTS) planning framework designed to enhance the controllability of LLM-driven dialogue agents. To enable this, we curate a dataset comprising SOP-annotated multi-scenario dialogues, generated using a semi-automated role-playing system with GPT-4o and validated through strict manual quality control. Additionally, we propose a novel method that integrates Chain of Thought reasoning with supervised fine-tuning for SOP prediction and utilizes SOP-guided Monte Carlo Tree Search for optimal action planning during dialogues. Experimental results demonstrate the effectiveness of our method, such as achieving a 27.95% improvement in action accuracy compared to baseline models based on GPT-3.5 and also showing notable gains for open-source models. Dataset and codes are publicly available.

Nithin Gopalakrishnan Nair, Anoop Cherian, Suhas Lohit et al.

Conditional generative models typically demand large annotated training sets to achieve high-quality synthesis. As a result, there has been significant interest in designing models that perform plug-and-play generation, i.e., to use a predefined or pretrained model, which is not explicitly trained on the generative task, to guide the generative process (e.g., using language). However, such guidance is typically useful only towards synthesizing high-level semantics rather than editing fine-grained details as in image-to-image translation tasks. To this end, and capitalizing on the powerful fine-grained generative control offered by the recent diffusion-based generative models, we introduce Steered Diffusion, a generalized framework for photorealistic zero-shot conditional image generation using a diffusion model trained for unconditional generation. The key idea is to steer the image generation of the diffusion model at inference time via designing a loss using a pre-trained inverse model that characterizes the conditional task. This loss modulates the sampling trajectory of the diffusion process. Our framework allows for easy incorporation of multiple conditions during inference. We present experiments using steered diffusion on several tasks including inpainting, colorization, text-guided semantic editing, and image super-resolution. Our results demonstrate clear qualitative and quantitative improvements over state-of-the-art diffusion-based plug-and-play models while adding negligible additional computational cost.

Ye Wang, Huazheng Pan, Tao Zhang et al.

The goal of document-level relation extraction (RE) is to identify relations between entities that span multiple sentences. Recently, incomplete labeling in document-level RE has received increasing attention, and some studies have used methods such as positive-unlabeled learning to tackle this issue, but there is still a lot of room for improvement. Motivated by this, we propose a positive-augmentation and positive-mixup positive-unlabeled metric learning framework (P3M). Specifically, we formulate document-level RE as a metric learning problem. We aim to pull the distance closer between entity pair embedding and their corresponding relation embedding, while pushing it farther away from the none-class relation embedding. Additionally, we adapt the positive-unlabeled learning to this loss objective. In order to improve the generalizability of the model, we use dropout to augment positive samples and propose a positive-none-class mixup method. Extensive experiments show that P3M improves the F1 score by approximately 4-10 points in document-level RE with incomplete labeling, and achieves state-of-the-art results in fully labeled scenarios. Furthermore, P3M has also demonstrated robustness to prior estimation bias in incomplete labeled scenarios.

Toshiaki Koike-Akino, Pu Wang, Ye Wang

Beyond data communications, commercial-off-the-shelf Wi-Fi devices can be used to monitor human activities, track device locomotion, and sense the ambient environment. In particular, spatial beam attributes that are inherently available in the 60-GHz IEEE 802.11ad/ay standards have shown to be effective in terms of overhead and channel measurement granularity for these indoor sensing tasks. In this paper, we investigate transfer learning to mitigate domain shift in human monitoring tasks when Wi-Fi settings and environments change over time. As a proof-of-concept study, we consider quantum neural networks (QNN) as well as classical deep neural networks (DNN) for the future quantum-ready society. The effectiveness of both DNN and QNN is validated by an in-house experiment for human pose recognition, achieving greater than 90% accuracy with a limited data size.

Xiangming Gu, Wei Zeng, Ye Wang

It is widely known that males and females typically possess different sound characteristics when singing, such as timbre and pitch, but it has never been explored whether these gender-based characteristics lead to a performance disparity in singing voice transcription (SVT), whose target includes pitch. Such a disparity could cause fairness issues and severely affect the user experience of downstream SVT applications. Motivated by this, we first demonstrate the female superiority of SVT systems, which is observed across different models and datasets. We find that different pitch distributions, rather than gender data imbalance, contribute to this disparity. To address this issue, we propose using an attribute predictor to predict gender labels and adversarially training the SVT system to enforce the gender-invariance of acoustic representations. Leveraging the prior knowledge that pitch distributions may contribute to the gender bias, we propose conditionally aligning acoustic representations between demographic groups by feeding note events to the attribute predictor. Empirical experiments on multiple benchmark SVT datasets show that our method significantly reduces gender bias (up to more than 50%) with negligible degradation of overall SVT performance, on both in-domain and out-of-domain singing data, thus offering a better fairness-utility trade-off.

Xiaodan Xi, Ge Li, Ye Wang et al.

We construct a strong PUF with provable security against ML attacks on both classical and quantum computers. The security is guaranteed by the cryptographic hardness of learning decryption functions of public-key cryptosystems, and the hardness of the learning-with-errors (LWE) problem defined on integer lattices. We call our construction the lattice PUF. We construct lattice PUF with a physically obfuscated key and an LWE decryption function block. To allow deployments in different scenarios, we demonstrate designs with different latency-area trade-offs. A compact design uses a highly serialized LFSR and LWE decryption function, while a latency-optimized design uses an unrolled LFSR and a parallel datapath. We prototype lattice PUF designs with $2^{136}$ challenge-response pairs (CRPs) on a Spartan 6 FPGA. In addition to theoretical security guarantee, we evaluate empirical resistance to the various leading ML techniques: the prediction error remains above $49.76\%$ after $1$ million training CRPs. The resource-efficient design requires only $45$ slices for the PUF logic proper, and $351$ slices for a fuzzy extractor. The latency-optimized design achieves a $148X$ reduction in latency, at a $10X$ increase in PUF hardware utilization. The mean uniformity of PUF responses is $49.98\%$, the mean uniqueness is $50.00\%$, and the mean reliability is $1.26\%$.

Yanmeng Wang, Jun Bai, Ye Wang et al.

Dual-Encoders is a promising mechanism for answer retrieval in question answering (QA) systems. Currently most conventional Dual-Encoders learn the semantic representations of questions and answers merely through matching score. Researchers proposed to introduce the QA interaction features in scoring function but at the cost of low efficiency in inference stage. To keep independent encoding of questions and answers during inference stage, variational auto-encoder is further introduced to reconstruct answers (questions) from question (answer) embeddings as an auxiliary task to enhance QA interaction in representation learning in training stage. However, the needs of text generation and answer retrieval are different, which leads to hardness in training. In this work, we propose a framework to enhance the Dual-Encoders model with question answer cross-embeddings and a novel Geometry Alignment Mechanism (GAM) to align the geometry of embeddings from Dual-Encoders with that from Cross-Encoders. Extensive experimental results show that our framework significantly improves Dual-Encoders model and outperforms the state-of-the-art method on multiple answer retrieval datasets.

Vivian Ziemke, Benjamin Estermann, Roger Wattenhofer et al.

In the evolving landscape of digital art, Non-Fungible Tokens (NFTs) have emerged as a groundbreaking platform, bridging the realms of art and technology. NFTs serve as the foundational framework that has revolutionized the market for digital art, enabling artists to showcase and monetize their creations in unprecedented ways. NFTs combine metadata stored on the blockchain with off-chain data, such as images, to create a novel form of digital ownership. It is not fully understood how these factors come together to determine NFT prices. In this study, we analyze both on-chain and off-chain data of NFT collections trading on OpenSea to understand what influences NFT pricing. Our results show that while text and image data of the NFTs can be used to explain price variations within collections, the extracted features do not generalize to new, unseen collections. Furthermore, we find that an NFT collection's trading volume often relates to its online presence, like social media followers and website traffic.

Jingwei Zhao, Gus Xia, Ziyu Wang et al.

In the realm of music AI, arranging rich and structured multi-track accompaniments from a simple lead sheet presents significant challenges. Such challenges include maintaining track cohesion, ensuring long-term coherence, and optimizing computational efficiency. In this paper, we introduce a novel system that leverages prior modelling over disentangled style factors to address these challenges. Our method presents a two-stage process: initially, a piano arrangement is derived from the lead sheet by retrieving piano texture styles; subsequently, a multi-track orchestration is generated by infusing orchestral function styles into the piano arrangement. Our key design is the use of vector quantization and a unique multi-stream Transformer to model the long-term flow of the orchestration style, which enables flexible, controllable, and structured music generation. Experiments show that by factorizing the arrangement task into interpretable sub-stages, our approach enhances generative capacity while improving efficiency. Additionally, our system supports a variety of music genres and provides style control at different composition hierarchies. We further show that our system achieves superior coherence, structure, and overall arrangement quality compared to existing baselines.

Md Rafi Ur Rashid, Jing Liu, Toshiaki Koike-Akino et al.

Fine-tuning large language models on private data for downstream applications poses significant privacy risks in potentially exposing sensitive information. Several popular community platforms now offer convenient distribution of a large variety of pre-trained models, allowing anyone to publish without rigorous verification. This scenario creates a privacy threat, as pre-trained models can be intentionally crafted to compromise the privacy of fine-tuning datasets. In this study, we introduce a novel poisoning technique that uses model-unlearning as an attack tool. This approach manipulates a pre-trained language model to increase the leakage of private data during the fine-tuning process. Our method enhances both membership inference and data extraction attacks while preserving model utility. Experimental results across different models, datasets, and fine-tuning setups demonstrate that our attacks significantly surpass baseline performance. This work serves as a cautionary note for users who download pre-trained models from unverified sources, highlighting the potential risks involved.

Kai Chen, Ye Wang, Yitong Li et al.

Temporal factors are tied to the growth of facts in realistic applications, such as the progress of diseases and the development of political situation, therefore, research on Temporal Knowledge Graph (TKG) attracks much attention. In TKG, relation patterns inherent with temporality are required to be studied for representation learning and reasoning across temporal facts. However, existing methods can hardly model temporal relation patterns, nor can capture the intrinsic connections between relations when evolving over time, lacking of interpretability. In this paper, we propose a novel temporal modeling method which represents temporal entities as Rotations in Quaternion Vector Space (RotateQVS) and relations as complex vectors in Hamilton's quaternion space. We demonstrate our method can model key patterns of relations in TKG, such as symmetry, asymmetry, inverse, and can further capture time-evolved relations by theory. Empirically, we show that our method can boost the performance of link prediction tasks over four temporal knowledge graph benchmarks.

Ye Wang, Yaxiong Wang, Guoshuai Zhao et al.

Few-shot class-incremental learning (FSCIL) aims to incrementally recognize new classes using a few samples while maintaining the performance on previously learned classes. One of the effective methods to solve this challenge is to construct prototypical evolution classifiers. Despite the advancement achieved by most existing methods, the classifier weights are simply initialized using mean features. Because representations for new classes are weak and biased, we argue such a strategy is suboptimal. In this paper, we tackle this issue from two aspects. Firstly, thanks to the development of foundation models, we employ a foundation model, the CLIP, as the network pedestal to provide a general representation for each class. Secondly, to generate a more reliable and comprehensive instance representation, we propose a Knowledge Adapter (KA) module that summarizes the data-specific knowledge from training data and fuses it into the general representation. Additionally, to tune the knowledge learned from the base classes to the upcoming classes, we propose a mechanism of Incremental Pseudo Episode Learning (IPEL) by simulating the actual FSCIL. Taken together, our proposed method, dubbed as Knowledge Adaptation Network (KANet), achieves competitive performance on a wide range of datasets, including CIFAR100, CUB200, and ImageNet-R.

Panfeng Cao, Ye Wang, Qiang Zhang et al.

Key information extraction (KIE) from scanned documents has gained increasing attention because of its applications in various domains. Although promising results have been achieved by some recent KIE approaches, they are usually built based on discriminative models, which lack the ability to handle optical character recognition (OCR) errors and require laborious token-level labelling. In this paper, we propose a novel generative end-to-end model, named GenKIE, to address the KIE task. GenKIE is a sequence-to-sequence multimodal generative model that utilizes multimodal encoders to embed visual, layout and textual features and a decoder to generate the desired output. Well-designed prompts are leveraged to incorporate the label semantics as the weakly supervised signals and entice the generation of the key information. One notable advantage of the generative model is that it enables automatic correction of OCR errors. Besides, token-level granular annotation is not required. Extensive experiments on multiple public real-world datasets show that GenKIE effectively generalizes over different types of documents and achieves state-of-the-art results. Our experiments also validate the model's robustness against OCR errors, making GenKIE highly applicable in real-world scenarios.

Hongfu Liu, Ye Wang

Large Language models (LLMs) possess the capability to engage In-context Learning (ICL) by leveraging a few demonstrations pertaining to a new downstream task as conditions. However, this particular learning paradigm suffers from high instability stemming from substantial variances induced by factors such as the input distribution of selected examples, their ordering, and prompt formats. In this work, we demonstrate that even when all these factors are held constant, the random selection of examples still results in high variance. Consequently, we aim to explore the informative ability of data examples by quantifying the Information Gain (IG) obtained in prediction after observing a given example candidate. Then we propose to sample those with maximum IG. Additionally, we identify the presence of template bias, which can lead to unfair evaluations of IG during the sampling process. To mitigate this bias, we introduce Calibration Before Sampling strategy. The experimental results illustrate that our proposed method can yield an average relative improvement of 14.3% across six classification tasks using three LLMs.

Hongfu Liu, Hengguan Huang, Ye Wang

Acoustic foundation models, fine-tuned for Automatic Speech Recognition (ASR), suffer from performance degradation in wild acoustic test settings when deployed in real-world scenarios. Stabilizing online Test-Time Adaptation (TTA) under these conditions remains an open and unexplored question. Existing wild vision TTA methods often fail to handle speech data effectively due to the unique characteristics of high-entropy speech frames, which are unreliably filtered out even when containing crucial semantic content. Furthermore, unlike static vision data, speech signals follow short-term consistency, requiring specialized adaptation strategies. In this work, we propose a novel wild acoustic TTA method tailored for ASR fine-tuned acoustic foundation models. Our method, Confidence-Enhanced Adaptation, performs frame-level adaptation using a confidence-aware weight scheme to avoid filtering out essential information in high-entropy frames. Additionally, we apply consistency regularization during test-time optimization to leverage the inherent short-term consistency of speech signals. Our experiments on both synthetic and real-world datasets demonstrate that our approach outperforms existing baselines under various wild acoustic test settings, including Gaussian noise, environmental sounds, accent variations, and sung speech.

Minjie Chen, Yao Cheng, Ye Wang et al.

Graph contrastive learning (GCL) has recently emerged as a promising approach for graph representation learning. Some existing methods adopt the 1-vs-K scheme to construct one positive and K negative samples for each graph, but it is difficult to set K. For those methods that do not use negative samples, it is often necessary to add additional strategies to avoid model collapse, which could only alleviate the problem to some extent. All these drawbacks will undoubtedly have an adverse impact on the generalizability and efficiency of the model. In this paper, to address these issues, we propose a novel graph self-contrast framework GraphSC, which only uses one positive and one negative sample, and chooses triplet loss as the objective. Specifically, self-contrast has two implications. First, GraphSC generates both positive and negative views of a graph sample from the graph itself via graph augmentation functions of various intensities, and use them for self-contrast. Second, GraphSC uses Hilbert-Schmidt Independence Criterion (HSIC) to factorize the representations into multiple factors and proposes a masked self-contrast mechanism to better separate positive and negative samples. Further, Since the triplet loss only optimizes the relative distance between the anchor and its positive/negative samples, it is difficult to ensure the absolute distance between the anchor and positive sample. Therefore, we explicitly reduced the absolute distance between the anchor and positive sample to accelerate convergence. Finally, we conduct extensive experiments to evaluate the performance of GraphSC against 19 other state-of-the-art methods in both unsupervised and transfer learning settings.

Longshen Ou, Xichu Ma, Ye Wang

Despite progress in melody-to-lyric generation, a substantial singability gap remains between machine-generated lyrics and those written by human lyricists. In this work, we aim to narrow this gap by jointly learning both wording and formatting for melody-to-lyric generation. After general-domain pretraining, our model acquires length awareness through an self-supervised stage trained on a large text-only lyric corpus. During supervised melody-to-lyric training, we introduce multiple auxiliary supervision objective informed by musicological findings on melody--lyric relationships, encouraging the model to capture fine-grained prosodic and structural patterns. Compared with naïve fine-tuning, our approach improves adherence to line-count and syllable-count requirements by 3.8% and 21.4% absolute, respectively, without degrading text quality. In human evaluation, it achieves 42.2% and 74.2% relative gains in overall quality over two task-specific baselines, underscoring the importance of formatting-aware training for generating singable lyrics.

Hengguan Huang, Songtao Wang, Hongfu Liu et al.

Traditional applications of natural language processing (NLP) in healthcare have predominantly focused on patient-centered services, enhancing patient interactions and care delivery, such as through medical dialogue systems. However, the potential of NLP to benefit inexperienced doctors, particularly in areas such as communicative medical coaching, remains largely unexplored. We introduce "ChatCoach", a human-AI cooperative framework designed to assist medical learners in practicing their communication skills during patient consultations. ChatCoach (Our data and code are available online: https://github.com/zerowst/Chatcoach)differentiates itself from conventional dialogue systems by offering a simulated environment where medical learners can practice dialogues with a patient agent, while a coach agent provides immediate, structured feedback. This is facilitated by our proposed Generalized Chain-of-Thought (GCoT) approach, which fosters the generation of structured feedback and enhances the utilization of external knowledge sources. Additionally, we have developed a dataset specifically for evaluating Large Language Models (LLMs) within the ChatCoach framework on communicative medical coaching tasks. Our empirical results validate the effectiveness of ChatCoach.

Ye Wang, Yaxiong Wang, Yujiao Wu et al.

Generalized Class Discovery (GCD) aims to dynamically assign labels to unlabelled data partially based on knowledge learned from labelled data, where the unlabelled data may come from known or novel classes. The prevailing approach generally involves clustering across all data and learning conceptions by prototypical contrastive learning. However, existing methods largely hinge on the performance of clustering algorithms and are thus subject to their inherent limitations. Firstly, the estimated cluster number is often smaller than the ground truth, making the existing methods suffer from the lack of prototypes for comprehensive conception learning. To address this issue, we propose an adaptive probing mechanism that introduces learnable potential prototypes to expand cluster prototypes (centers). As there is no ground truth for the potential prototype, we develop a self-supervised prototype learning framework to optimize the potential prototype in an end-to-end fashion. Secondly, clustering is computationally intensive, and the conventional strategy of clustering both labelled and unlabelled instances exacerbates this issue. To counteract this inefficiency, we opt to cluster only the unlabelled instances and subsequently expand the cluster prototypes with our introduced potential prototypes to fast explore novel classes. Despite the simplicity of our proposed method, extensive empirical analysis on a wide range of datasets confirms that our method consistently delivers state-of-the-art results. Specifically, our method surpasses the nearest competitor by a significant margin of 9.7% within the Stanford Cars dataset and 12x clustering efficiency within the Herbarium 19 dataset. We will make the code and checkpoints publicly available at https://github.com/xjtuYW/PNP.git.

Wei Wei, Huang Hengguan, Gu Xiangming et al.

Content mismatch usually occurs when data from one modality is translated to another, e.g. language learners producing mispronunciations (errors in speech) when reading a sentence (target text) aloud. However, most existing alignment algorithms assume that the content involved in the two modalities is perfectly matched, thus leading to difficulty in locating such mismatch between speech and text. In this work, we develop an unsupervised learning algorithm that can infer the relationship between content-mismatched cross-modal sequential data, especially for speech-text sequences. More specifically, we propose a hierarchical Bayesian deep learning model, dubbed mismatch localization variational autoencoder (ML-VAE), which decomposes the generative process of the speech into hierarchically structured latent variables, indicating the relationship between the two modalities. Training such a model is very challenging due to the discrete latent variables with complex dependencies involved. To address this challenge, we propose a novel and effective training procedure that alternates between estimating the hard assignments of the discrete latent variables over a specifically designed mismatch localization finite-state acceptor (ML-FSA) and updating the parameters of neural networks. In this work, we focus on the mismatch localization problem for speech and text, and our experimental results show that ML-VAE successfully locates the mismatch between text and speech, without the need for human annotations for model training.

Longshen Ou, Xichu Ma, Min-Yen Kan et al.

The development of general-domain neural machine translation (NMT) methods has advanced significantly in recent years, but the lack of naturalness and musical constraints in the outputs makes them unable to produce singable lyric translations. This paper bridges the singability quality gap by formalizing lyric translation into a constrained translation problem, converting theoretical guidance and practical techniques from translatology literature to prompt-driven NMT approaches, exploring better adaptation methods, and instantiating them to an English-Chinese lyric translation system. Our model achieves 99.85%, 99.00%, and 95.52% on length accuracy, rhyme accuracy, and word boundary recall. In our subjective evaluation, our model shows a 75% relative enhancement on overall quality, compared against naive fine-tuning (Code available at https://github.com/Sonata165/ControllableLyricTranslation).

Sai Samarth R Phaye, Emmanouil Benetos, Ye Wang

Acoustic Scene Classification (ASC) is one of the core research problems in the field of Computational Sound Scene Analysis. In this work, we present SubSpectralNet, a novel model which captures discriminative features by incorporating frequency band-level differences to model soundscapes. Using mel-spectrograms, we propose the idea of using band-wise crops of the input time-frequency representations and train a convolutional neural network (CNN) on the same. We also propose a modification in the training method for more efficient learning of the CNN models. We first give a motivation for using sub-spectrograms by giving intuitive and statistical analyses and finally we develop a sub-spectrogram based CNN architecture for ASC. The system is evaluated on the public ASC development dataset provided for the "Detection and Classification of Acoustic Scenes and Events" (DCASE) 2018 Challenge. Our best model achieves an improvement of +14% in terms of classification accuracy with respect to the DCASE 2018 baseline system. Code and figures are available at https://github.com/ssrp/SubSpectralNet

Xiangming Gu, Xiaosen Zheng, Tianyu Pang et al. · tsinghua

A multimodal large language model (MLLM) agent can receive instructions, capture images, retrieve histories from memory, and decide which tools to use. Nonetheless, red-teaming efforts have revealed that adversarial images/prompts can jailbreak an MLLM and cause unaligned behaviors. In this work, we report an even more severe safety issue in multi-agent environments, referred to as infectious jailbreak. It entails the adversary simply jailbreaking a single agent, and without any further intervention from the adversary, (almost) all agents will become infected exponentially fast and exhibit harmful behaviors. To validate the feasibility of infectious jailbreak, we simulate multi-agent environments containing up to one million LLaVA-1.5 agents, and employ randomized pair-wise chat as a proof-of-concept instantiation for multi-agent interaction. Our results show that feeding an (infectious) adversarial image into the memory of any randomly chosen agent is sufficient to achieve infectious jailbreak. Finally, we derive a simple principle for determining whether a defense mechanism can provably restrain the spread of infectious jailbreak, but how to design a practical defense that meets this principle remains an open question to investigate. Our project page is available at https://sail-sg.github.io/Agent-Smith/.

Yan Cai, Linlin Wang, Ye Wang et al.

The emergence of various medical large language models (LLMs) in the medical domain has highlighted the need for unified evaluation standards, as manual evaluation of LLMs proves to be time-consuming and labor-intensive. To address this issue, we introduce MedBench, a comprehensive benchmark for the Chinese medical domain, comprising 40,041 questions sourced from authentic examination exercises and medical reports of diverse branches of medicine. In particular, this benchmark is composed of four key components: the Chinese Medical Licensing Examination, the Resident Standardization Training Examination, the Doctor In-Charge Qualification Examination, and real-world clinic cases encompassing examinations, diagnoses, and treatments. MedBench replicates the educational progression and clinical practice experiences of doctors in Mainland China, thereby establishing itself as a credible benchmark for assessing the mastery of knowledge and reasoning abilities in medical language learning models. We perform extensive experiments and conduct an in-depth analysis from diverse perspectives, which culminate in the following findings: (1) Chinese medical LLMs underperform on this benchmark, highlighting the need for significant advances in clinical knowledge and diagnostic precision. (2) Several general-domain LLMs surprisingly possess considerable medical knowledge. These findings elucidate both the capabilities and limitations of LLMs within the context of MedBench, with the ultimate goal of aiding the medical research community.

Andrew Lowy, Zhuohang Li, Jing Liu et al.

For small privacy parameter $ε$, $ε$-differential privacy (DP) provides a strong worst-case guarantee that no membership inference attack (MIA) can succeed at determining whether a person's data was used to train a machine learning model. The guarantee of DP is worst-case because: a) it holds even if the attacker already knows the records of all but one person in the data set; and b) it holds uniformly over all data sets. In practical applications, such a worst-case guarantee may be overkill: practical attackers may lack exact knowledge of (nearly all of) the private data, and our data set might be easier to defend, in some sense, than the worst-case data set. Such considerations have motivated the industrial deployment of DP models with large privacy parameter (e.g. $ε\geq 7$), and it has been observed empirically that DP with large $ε$ can successfully defend against state-of-the-art MIAs. Existing DP theory cannot explain these empirical findings: e.g., the theoretical privacy guarantees of $ε\geq 7$ are essentially vacuous. In this paper, we aim to close this gap between theory and practice and understand why a large DP parameter can prevent practical MIAs. To tackle this problem, we propose a new privacy notion called practical membership privacy (PMP). PMP models a practical attacker's uncertainty about the contents of the private data. The PMP parameter has a natural interpretation in terms of the success rate of a practical MIA on a given data set. We quantitatively analyze the PMP parameter of two fundamental DP mechanisms: the exponential mechanism and Gaussian mechanism. Our analysis reveals that a large DP parameter often translates into a much smaller PMP parameter, which guarantees strong privacy against practical MIAs. Using our findings, we offer principled guidance for practitioners in choosing the DP parameter.

Ye Wang, Xinrun Xu, Zhiming Ding

When completing knowledge-intensive tasks, humans sometimes need an answer and a corresponding reference passage for auxiliary reading. Previous methods required obtaining pre-segmented article chunks through additional retrieval models. This paper explores leveraging the parameterized knowledge stored during the pre-training phase of large language models (LLMs) to recall reference passage from any starting position independently. We propose a two-stage framework that simulates the scenario of humans recalling easily forgotten references. Initially, the LLM is prompted to recall document title identifiers to obtain a coarse-grained document set. Then, based on the acquired coarse-grained document set, it recalls fine-grained passage. In the two-stage recall process, we use constrained decoding to ensure that content outside of the stored documents is not generated. To increase speed, we only recall a short prefix in the second stage, and then locate its position to retrieve a complete passage. Experiments on KILT knowledge-sensitive tasks have verified that LLMs can independently recall reference passage locations in various task forms, and the obtained reference significantly assists downstream tasks.

Md Rafi Ur Rashid, Vishnu Asutosh Dasu, Ye Wang et al.

Large Language Models (LLMs) exhibit impressive capabilities, but remain susceptible to a growing spectrum of safety risks, including jailbreaks, toxic content, hallucinations, and bias. Existing defenses often address only a single threat type or resort to rigid outright rejection, sacrificing user experience and failing to generalize across diverse and novel attacks. This paper introduces Adversarial Scenario Extrapolation (ASE), a novel inference-time computation framework that leverages Chain-of-Thought (CoT) reasoning to simultaneously enhance LLM robustness and seamlessness. ASE guides the LLM through a self-generative process of contemplating potential adversarial scenarios and formulating defensive strategies before generating a response to the user query. Comprehensive evaluation on four adversarial benchmarks with four latest LLMs shows that ASE achieves near-zero jailbreak attack success rates and minimal toxicity, while slashing outright rejections to <4%. ASE outperforms six state-of-the-art defenses in robustness-seamlessness trade-offs, with 92-99% accuracy on adversarial Q&A and 4-10x lower bias scores. By transforming adversarial perception into an intrinsic cognitive process, ASE sets a new paradigm for secure and natural human-AI interaction.

Yuting Mei, Ye Wang, Sipeng Zheng et al.

As robotic agents increasingly assist humans in reality, quadruped robots offer unique opportunities for interaction in complex scenarios due to their agile movement. However, building agents that can autonomously navigate, adapt, and respond to versatile goals remains a significant challenge. In this work, we introduce QuadrupedGPT designed to follow diverse commands with agility comparable to that of a pet. The primary challenges addressed include: i) effectively utilizing multimodal observations for informed decision-making; ii) achieving agile control by integrating locomotion and navigation; iii) developing advanced cognition to execute long-term objectives. Our QuadrupedGPT interprets human commands and environmental contexts using a large multimodal model. Leveraging its extensive knowledge base, the agent autonomously assigns parameters for adaptive locomotion policies and devises safe yet efficient paths toward its goals. Additionally, it employs high-level reasoning to decompose long-term goals into a sequence of executable subgoals. Through comprehensive experiments, our agent shows proficiency in handling diverse tasks and intricate instructions, representing a significant step toward the development of versatile quadruped agents for open-ended environments.

Xintong Wang, Mingqian Shi, Ye Wang

Mispronunciation Detection and Diagnosis (MDD) systems, leveraging Automatic Speech Recognition (ASR), face two main challenges in Mandarin Chinese: 1) The two-stage models create an information gap between the phoneme or tone classification stage and the MDD stage. 2) The scarcity of Mandarin MDD datasets limits model training. In this paper, we introduce a stateless RNN-T model for Mandarin MDD, utilizing HuBERT features with pitch embedding through a Pitch Fusion Block. Our model, trained solely on native speaker data, shows a 3% improvement in Phone Error Rate and a 7% increase in False Acceptance Rate over the state-of-the-art baseline in non-native scenarios

Ye Wang, Zili Yi, Rui Ma

Personalized text-to-image (T2I) models not only produce lifelike and varied visuals but also allow users to tailor the images to fit their personal taste. These personalization techniques can grasp the essence of a concept through a collection of images, or adjust a pre-trained text-to-image model with a specific image input for subject-driven or attribute-aware guidance. Yet, accurately capturing the distinct visual attributes of an individual image poses a challenge for these methods. To address this issue, we introduce OSTAF, a novel parameter-efficient one-shot fine-tuning method which only utilizes one reference image for T2I personalization. A novel hypernetwork-powered attribute-focused fine-tuning mechanism is employed to achieve the precise learning of various attribute features (e.g., appearance, shape or drawing style) from the reference image. Comparing to existing image customization methods, our method shows significant superiority in attribute identification and application, as well as achieves a good balance between efficiency and output quality.

Xu Yang, Jiyuan Feng, Songyue Guo et al.

Personalized federated learning becomes a hot research topic that can learn a personalized learning model for each client. Existing personalized federated learning models prefer to aggregate similar clients with similar data distribution to improve the performance of learning models. However, similaritybased personalized federated learning methods may exacerbate the class imbalanced problem. In this paper, we propose a novel Dynamic Affinity-based Personalized Federated Learning model (DA-PFL) to alleviate the class imbalanced problem during federated learning. Specifically, we build an affinity metric from a complementary perspective to guide which clients should be aggregated. Then we design a dynamic aggregation strategy to dynamically aggregate clients based on the affinity metric in each round to reduce the class imbalanced risk. Extensive experiments show that the proposed DA-PFL model can significantly improve the accuracy of each client in three real-world datasets with state-of-the-art comparison methods.

Sipeng Zheng, Bohan Zhou, Yicheng Feng et al.

In this paper, we propose \textbf{UniCode}, a novel approach within the domain of multimodal large language models (MLLMs) that learns a unified codebook to efficiently tokenize visual, text, and potentially other types of signals. This innovation addresses a critical limitation in existing MLLMs: their reliance on a text-only codebook, which restricts MLLM's ability to generate images and texts in a multimodal context. Towards this end, we propose a language-driven iterative training paradigm, coupled with an in-context pre-training task we term ``image decompression'', enabling our model to interpret compressed visual data and generate high-quality images.The unified codebook empowers our model to extend visual instruction tuning to non-linguistic generation tasks. Moreover, UniCode is adaptable to diverse stacked quantization approaches in order to compress visual signals into a more compact token representation. Despite using significantly fewer parameters and less data during training, Unicode demonstrates promising capabilities in visual reconstruction and generation. It also achieves performances comparable to leading MLLMs across a spectrum of VQA benchmarks.

Hongfu Liu, Mingqian Shi, Ye Wang

Automatic Pronunciation Assessment (APA) is vital for computer-assisted language learning. Prior methods rely on annotated speech-text data to train Automatic Speech Recognition (ASR) models or speech-score data to train regression models. In this work, we propose a novel zero-shot APA method based on the pre-trained acoustic model, HuBERT. Our method involves encoding speech input and corrupting them via a masking module. We then employ the Transformer encoder and apply k-means clustering to obtain token sequences. Finally, a scoring module is designed to measure the number of wrongly recovered tokens. Experimental results on speechocean762 demonstrate that the proposed method achieves comparable performance to supervised regression baselines and outperforms non-regression baselines in terms of Pearson Correlation Coefficient (PCC). Additionally, we analyze how masking strategies affect the performance of APA.

Safa C. Medin, Bernhard Egger, Anoop Cherian et al.

Recent advances in generative adversarial networks (GANs) have led to remarkable achievements in face image synthesis. While methods that use style-based GANs can generate strikingly photorealistic face images, it is often difficult to control the characteristics of the generated faces in a meaningful and disentangled way. Prior approaches aim to achieve such semantic control and disentanglement within the latent space of a previously trained GAN. In contrast, we propose a framework that a priori models physical attributes of the face such as 3D shape, albedo, pose, and lighting explicitly, thus providing disentanglement by design. Our method, MOST-GAN, integrates the expressive power and photorealism of style-based GANs with the physical disentanglement and flexibility of nonlinear 3D morphable models, which we couple with a state-of-the-art 2D hair manipulation network. MOST-GAN achieves photorealistic manipulation of portrait images with fully disentangled 3D control over their physical attributes, enabling extreme manipulation of lighting, facial expression, and pose variations up to full profile view.

Aditya Sanghi, Hang Chu, Joseph G. Lambourne et al.

Generating shapes using natural language can enable new ways of imagining and creating the things around us. While significant recent progress has been made in text-to-image generation, text-to-shape generation remains a challenging problem due to the unavailability of paired text and shape data at a large scale. We present a simple yet effective method for zero-shot text-to-shape generation that circumvents such data scarcity. Our proposed method, named CLIP-Forge, is based on a two-stage training process, which only depends on an unlabelled shape dataset and a pre-trained image-text network such as CLIP. Our method has the benefits of avoiding expensive inference time optimization, as well as the ability to generate multiple shapes for a given text. We not only demonstrate promising zero-shot generalization of the CLIP-Forge model qualitatively and quantitatively, but also provide extensive comparative evaluations to better understand its behavior.

Qianmengke Zhao, Ye Wang, Qun Liu

Although deep learning models are powerful among various applications, most deep learning models are still a black box, lacking verifiability and interpretability, which means the decision-making process that human beings cannot understand. Therefore, how to evaluate deep neural networks with explanations is still an urgent task. In this paper, we first propose a multi-semantic image recognition model, which enables human beings to understand the decision-making process of the neural network. Then, we presents a new evaluation index, which can quantitatively assess the model interpretability. We also comprehensively summarize the semantic information that affects the image classification results in the judgment process of neural networks. Finally, this paper also exhibits the relevant baseline performance with current state-of-the-art deep learning models.

Hengguan Huang, Hongfu Liu, Hao Wang et al.

Perception of time from sequentially acquired sensory inputs is rooted in everyday behaviors of individual organisms. Yet, most algorithms for time-series modeling fail to learn dynamics of random event timings directly from visual or audio inputs, requiring timing annotations during training that are usually unavailable for real-world applications. For instance, neuroscience perspectives on postdiction imply that there exist variable temporal ranges within which the incoming sensory inputs can affect the earlier perception, but such temporal ranges are mostly unannotated for real applications such as automatic speech recognition (ASR). In this paper, we present a probabilistic ordinary differential equation (ODE), called STochastic boundaRy ODE (STRODE), that learns both the timings and the dynamics of time series data without requiring any timing annotations during training. STRODE allows the usage of differential equations to sample from the posterior point processes, efficiently and analytically. We further provide theoretical guarantees on the learning of STRODE. Our empirical results show that our approach successfully infers event timings of time series data. Our method achieves competitive or superior performances compared to existing state-of-the-art methods for both synthetic and real-world datasets.

Shuqi Dai, Xichu Ma, Ye Wang et al.

Many practices have been presented in music generation recently. While stylistic music generation using deep learning techniques has became the main stream, these models still struggle to generate music with high musicality, different levels of music structure, and controllability. In addition, more application scenarios such as music therapy require imitating more specific musical styles from a few given music examples, rather than capturing the overall genre style of a large data corpus. To address requirements that challenge current deep learning methods, we propose a statistical machine learning model that is able to capture and imitate the structure, melody, chord, and bass style from a given example seed song. An evaluation using 10 pop songs shows that our new representations and methods are able to create high-quality stylistic music that is similar to a given input song. We also discuss potential uses of our approach in music evaluation and music therapy.

Kaitai Zhang, Bin Wang, Hong-Shuo Chen et al.

The main challenge of dynamic texture synthesis lies in how to maintain spatial and temporal consistency in synthesized videos. The major drawback of existing dynamic texture synthesis models comes from poor treatment of the long-range texture correlation and motion information. To address this problem, we incorporate a new loss term, called the Shifted Gram loss, to capture the structural and long-range correlation of the reference texture video. Furthermore, we introduce a frame sampling strategy to exploit long-period motion across multiple frames. With these two new techniques, the application scope of existing texture synthesis models can be extended. That is, they can synthesize not only homogeneous but also structured dynamic texture patterns. Thorough experimental results are provided to demonstrate that our proposed dynamic texture synthesis model offers state-of-the-art visual performance.

Jialin Peng, Ye Wang

Despite the remarkable performance of deep learning methods on various tasks, most cutting-edge models rely heavily on large-scale annotated training examples, which are often unavailable for clinical and health care tasks. The labeling costs for medical images are very high, especially in medical image segmentation, which typically requires intensive pixel/voxel-wise labeling. Therefore, the strong capability of learning and generalizing from limited supervision, including a limited amount of annotations, sparse annotations, and inaccurate annotations, is crucial for the successful application of deep learning models in medical image segmentation. However, due to its intrinsic difficulty, segmentation with limited supervision is challenging and specific model design and/or learning strategies are needed. In this paper, we provide a systematic and up-to-date review of the solutions above, with summaries and comments about the methodologies. We also highlight several problems in this field, discussed future directions observing further investigations.

Wenjie Chen, Fengtong Du, Ye Wang et al.

The problem of catastrophic forgetting has a history of more than 30 years and has not been completely solved yet. Since the human brain has natural ability to perform continual lifelong learning, learning from the brain may provide solutions to this problem. In this paper, we propose a novel biologically plausible audio-visual integration model (AVIM) based on the assumption that the integration of audio and visual perceptual information in the medial temporal lobe during learning is crucial to form concepts and make continual learning possible. Specifically, we use multi-compartment Hodgkin-Huxley neurons to build the model and adopt the calcium-based synaptic tagging and capture as the model's learning rule. Furthermore, we define a new continual learning paradigm to simulate the possible continual learning process in the human brain. We then test our model under this new paradigm. Our experimental results show that the proposed AVIM can achieve state-of-the-art continual learning performance compared with other advanced methods such as OWM, iCaRL and GEM. Moreover, it can generate stable representations of objects during learning. These results support our assumption that concept formation is essential for continuous lifelong learning and suggest the proposed AVIM is a possible concept formation mechanism.

Hengguan Huang, Fuzhao Xue, Hao Wang et al.

Lying at the core of human intelligence, relational thinking is characterized by initially relying on innumerable unconscious percepts pertaining to relations between new sensory signals and prior knowledge, consequently becoming a recognizable concept or object through coupling and transformation of these percepts. Such mental processes are difficult to model in real-world problems such as in conversational automatic speech recognition (ASR), as the percepts (if they are modelled as graphs indicating relationships among utterances) are supposed to be innumerable and not directly observable. In this paper, we present a Bayesian nonparametric deep learning method called deep graph random process (DGP) that can generate an infinite number of probabilistic graphs representing percepts. We further provide a closed-form solution for coupling and transformation of these percept graphs for acoustic modeling. Our approach is able to successfully infer relations among utterances without using any relational data during training. Experimental evaluations on ASR tasks including CHiME-2 and CHiME-5 demonstrate the effectiveness and benefits of our method.

Toshiaki Koike-Akino, Ye Wang

We propose a new concept of rateless auto-encoders (RL-AEs) that enable a flexible latent dimensionality, which can be seamlessly adjusted for varying distortion and dimensionality requirements. In the proposed RL-AEs, instead of a deterministic bottleneck architecture, we use an over-complete representation that is stochastically regularized with weighted dropouts, in a manner analogous to sparse AE (SAE). Unlike SAEs, our RL-AEs employ monotonically increasing dropout rates across the latent representation nodes such that the latent variables become sorted by importance like in principal component analysis (PCA). This is motivated by the rateless property of conventional PCA, where the least important principal components can be discarded to realize variable rate dimensionality reduction that gracefully degrades the distortion. In contrast, since the latent variables of conventional AEs are equally important for data reconstruction, they cannot be simply discarded to further reduce the dimensionality after the AE model is trained. Our proposed stochastic bottleneck framework enables seamless rate adaptation with high reconstruction performance, without requiring predetermined latent dimensionality at training. We experimentally demonstrate that the proposed RL-AEs can achieve variable dimensionality reduction while achieving low distortion compared to conventional AEs.