Current audio-visual representation learning can capture rough object categories (e.g., ``animals'' and ``instruments''), but it lacks the ability to recognize fine-grained details, such as specific categories like ``dogs'' and ``flutes'' within animals and instruments. To address this issue, we introduce DETECLAP, a method to enhance audio-visual representation learning with object information. Our key idea is to introduce an audio-visual label prediction loss to the existing Contrastive Audio-Visual Masked AutoEncoder to enhance its object awareness. To avoid costly manual annotations, we prepare object labels from both audio and visual inputs using state-of-the-art language-audio models and object detectors. We evaluate the method of audio-visual retrieval and classification using the VGGSound and AudioSet20K datasets. Our method achieves improvements in recall@10 of +1.5% and +1.2% for audio-to-visual and visual-to-audio retrieval, respectively, and an improvement in accuracy of +0.6% for audio-visual classification.
Audio-text contrastive models have become a powerful approach in music representation learning. Despite their empirical success, however, little is known about the influence of key design choices on the quality of music-text representations learnt through this framework. In this work, we expose these design choices within the constraints of limited data and computation budgets, and establish a more solid understanding of their impact grounded in empirical observations along three axes: the choice of base encoders, the level of curation in training data, and the use of text augmentation. We find that data curation is the single most important factor for music-text contrastive training in resource-constrained scenarios. Motivated by this insight, we introduce two novel techniques, Augmented View Dropout and TextSwap, which increase the diversity and descriptiveness of text inputs seen in training. Through our experiments we demonstrate that these are effective at boosting performance across different pre-training regimes, model architectures, and downstream data distributions, without incurring higher computational costs or requiring additional training data.
Robotic assistive feeding holds significant promise for improving the quality of life for individuals with eating disabilities. However, acquiring diverse food items under varying conditions and generalizing to unseen food presents unique challenges. Existing methods that rely on surface-level geometric information (e.g., bounding box and pose) derived from visual cues (e.g., color, shape, and texture) often lacks adaptability and robustness, especially when foods share similar physical properties but differ in visual appearance. We employ imitation learning (IL) to learn a policy for food acquisition. Existing methods employ IL or Reinforcement Learning (RL) to learn a policy based on off-the-shelf image encoders such as ResNet-50. However, such representations are not robust and struggle to generalize across diverse acquisition scenarios. To address these limitations, we propose a novel approach, IMRL (Integrated Multi-Dimensional Representation Learning), which integrates visual, physical, temporal, and geometric representations to enhance the robustness and generalizability of IL for food acquisition. Our approach captures food types and physical properties (e.g., solid, semi-solid, granular, liquid, and mixture), models temporal dynamics of acquisition actions, and introduces geometric information to determine optimal scooping points and assess bowl fullness. IMRL enables IL to adaptively adjust scooping strategies based on context, improving the robot's capability to handle diverse food acquisition scenarios. Experiments on a real robot demonstrate our approach's robustness and adaptability across various foods and bowl configurations, including zero-shot generalization to unseen settings. Our approach achieves improvement up to $35\%$ in success rate compared with the best-performing baseline. More details can be found on our website https://ruiiu.github.io/imrl.
Multimodal schizophrenia assessment systems have gained traction over the last few years. This work introduces a schizophrenia assessment system to discern between prominent symptom classes of schizophrenia and predict an overall schizophrenia severity score. We develop a Vector Quantized Variational Auto-Encoder (VQ-VAE) based Multimodal Representation Learning (MRL) model to produce task-agnostic speech representations from vocal Tract Variables (TVs) and Facial Action Units (FAUs). These representations are then used in a Multi-Task Learning (MTL) based downstream prediction model to obtain class labels and an overall severity score. The proposed framework outperforms the previous works on the multi-class classification task across all evaluation metrics (Weighted F1 score, AUC-ROC score, and Weighted Accuracy). Additionally, it estimates the schizophrenia severity score, a task not addressed by earlier approaches.
Self-supervised speech representation learning has become essential for extracting meaningful features from untranscribed audio. Recent advances highlight the potential of deriving discrete symbols from the features correlated with linguistic units, which enables text-less training across diverse tasks. In particular, sentence-level Self-Distillation of the pretrained HuBERT (SD-HuBERT) induces syllabic structures within latent speech frame representations extracted from an intermediate Transformer layer. In SD-HuBERT, sentence-level representation is accumulated from speech frame features through self-attention layers using a special CLS token. However, we observe that the information aggregated in the CLS token correlates more with speaker identity than with linguistic content. To address this, we propose a speech-only self-supervised fine-tuning approach that separates syllabic units from speaker information. Our method introduces speaker perturbation as data augmentation and adopts a frame-level training objective to prevent the CLS token from aggregating paralinguistic information. Experimental results show that our approach surpasses the current state-of-the-art method in most syllable segmentation and syllabic unit quality metrics on Librispeech, underscoring its effectiveness in promoting syllabic organization within speech-only models.
We introduce AnnualBERT, a series of language models designed specifically to capture the temporal evolution of scientific text. Deviating from the prevailing paradigms of subword tokenizations and "one model to rule them all", AnnualBERT adopts whole words as tokens and is composed of a base RoBERTa model pretrained from scratch on the full-text of 1.7 million arXiv papers published until 2008 and a collection of progressively trained models on arXiv papers at an annual basis. We demonstrate the effectiveness of AnnualBERT models by showing that they not only have comparable performances in standard tasks but also achieve state-of-the-art performances on domain-specific NLP tasks as well as link prediction tasks in the arXiv citation network. We then utilize probing tasks to quantify the models' behavior in terms of representation learning and forgetting as time progresses. Our approach enables the pretrained models to not only improve performances on scientific text processing tasks but also to provide insights into the development of scientific discourse over time. The series of the models is available at https://huggingface.co/jd445/AnnualBERTs.
Contrastive learning has become one of the most impressive approaches for multi-modal representation learning. However, previous multi-modal works mainly focused on cross-modal understanding, ignoring in-modal contrastive learning, which limits the representation of each modality. In this paper, we propose a novel contrastive learning strategy, called $Turbo$, to promote multi-modal understanding by joint in-modal and cross-modal contrastive learning. Specifically, multi-modal data pairs are sent through the forward pass twice with different hidden dropout masks to get two different representations for each modality. With these representations, we obtain multiple in-modal and cross-modal contrastive objectives for training. Finally, we combine the self-supervised Turbo with the supervised multi-modal classification and demonstrate its effectiveness on two audio-text classification tasks, where the state-of-the-art performance is achieved on a speech emotion recognition benchmark dataset.
Graph pooling compresses graphs and summarises their topological properties and features in a vectorial representation. It is an essential part of deep graph representation learning and is indispensable in graph-level tasks like classification or regression. Current approaches pool hierarchical structures in graphs by iteratively applying shallow pooling operators up to a fixed depth. However, they disregard the interdependencies between structures at different hierarchical levels and do not adapt to datasets that contain graphs with different sizes that may require pooling with various depths. To address these issues, we propose MDL-Pool, a pooling operator based on the minimum description length (MDL) principle, whose loss formulation explicitly models the interdependencies between different hierarchical levels and facilitates a direct comparison between multiple pooling alternatives with different depths. MDP-Pool builds on the map equation, an information-theoretic objective function for community detection, which naturally implements Occam's razor and balances between model complexity and goodness-of-fit via the MDL. We demonstrate MDL-Pool's competitive performance in an empirical evaluation against various baselines across standard graph classification datasets.
Current end-to-end autonomous driving methods resort to unifying modular designs for various tasks (e.g. perception, prediction and planning). Although optimized in a planning-oriented spirit with a fully differentiable framework, existing end-to-end driving systems without ego-centric designs still suffer from unsatisfactory performance and inferior efficiency, owing to the rasterized scene representation learning and redundant information transmission. In this paper, we revisit the human driving behavior and propose an ego-centric fully sparse paradigm, named DiFSD, for end-to-end self-driving. Specifically, DiFSD mainly consists of sparse perception, hierarchical interaction and iterative motion planner. The sparse perception module performs detection, tracking and online mapping based on sparse representation of the driving scene. The hierarchical interaction module aims to select the Closest In-Path Vehicle / Stationary (CIPV / CIPS) from coarse to fine, benefiting from an additional geometric prior. As for the iterative motion planner, both selected interactive agents and ego-vehicle are considered for joint motion prediction, where the output multi-modal ego-trajectories are optimized in an iterative fashion. Besides, both position-level motion diffusion and trajectory-level planning denoising are introduced for uncertainty modeling, thus facilitating the training stability and convergence of the whole framework. Extensive experiments conducted on nuScenes and Bench2Drive datasets demonstrate the superior planning performance and great efficiency of DiFSD.
The rapid development of photo-realistic face generation methods has raised significant concerns in society and academia, highlighting the urgent need for robust and generalizable face forgery detection (FFD) techniques. Although existing approaches mainly capture face forgery patterns using image modality, other modalities like fine-grained noises and texts are not fully explored, which limits the generalization capability of the model. In addition, most FFD methods tend to identify facial images generated by GAN, but struggle to detect unseen diffusion-synthesized ones. To address the limitations, we aim to leverage the cutting-edge foundation model, contrastive language-image pre-training (CLIP), to achieve generalizable diffusion face forgery detection (DFFD). In this paper, we propose a novel multi-modal fine-grained CLIP (MFCLIP) model, which mines comprehensive and fine-grained forgery traces across image-noise modalities via language-guided face forgery representation learning, to facilitate the advancement of DFFD. Specifically, we devise a fine-grained language encoder (FLE) that extracts fine global language features from hierarchical text prompts. We design a multi-modal vision encoder (MVE) to capture global image forgery embeddings as well as fine-grained noise forgery patterns extracted from the richest patch, and integrate them to mine general visual forgery traces. Moreover, we build an innovative plug-and-play sample pair attention (SPA) method to emphasize relevant negative pairs and suppress irrelevant ones, allowing cross-modality sample pairs to conduct more flexible alignment. Extensive experiments and visualizations show that our model outperforms the state of the arts on different settings like cross-generator, cross-forgery, and cross-dataset evaluations.
While motion has garnered attention in various tasks, its potential as a modality for weakly-supervised object detection (WSOD) in static images remains unexplored. Our study introduces an approach to enhance WSOD methods by integrating motion information. This method involves leveraging hallucinated motion from static images to improve WSOD on image datasets, utilizing a Siamese network for enhanced representation learning with motion, addressing camera motion through motion normalization, and selectively training images based on object motion. Experimental validation on the COCO and YouTube-BB datasets demonstrates improvements over a state-of-the-art method.
Self-supervised learning has proved effective for skeleton-based human action understanding. However, previous works either rely on contrastive learning that suffers false negative problems or are based on reconstruction that learns too much unessential low-level clues, leading to limited representations for downstream tasks. Recently, great advances have been made in generative learning, which is naturally a challenging yet meaningful pretext task to model the general underlying data distributions. However, the representation learning capacity of generative models is under-explored, especially for the skeletons with spacial sparsity and temporal redundancy. To this end, we propose Masked Conditional Diffusion (MacDiff) as a unified framework for human skeleton modeling. For the first time, we leverage diffusion models as effective skeleton representation learners. Specifically, we train a diffusion decoder conditioned on the representations extracted by a semantic encoder. Random masking is applied to encoder inputs to introduce a information bottleneck and remove redundancy of skeletons. Furthermore, we theoretically demonstrate that our generative objective involves the contrastive learning objective which aligns the masked and noisy views. Meanwhile, it also enforces the representation to complement for the noisy view, leading to better generalization performance. MacDiff achieves state-of-the-art performance on representation learning benchmarks while maintaining the competence for generative tasks. Moreover, we leverage the diffusion model for data augmentation, significantly enhancing the fine-tuning performance in scenarios with scarce labeled data. Our project is available at https://lehongwu.github.io/ECCV24MacDiff/.
We introduce jina-embeddings-v3, a novel text embedding model with 570 million parameters, achieves state-of-the-art performance on multilingual data and long-context retrieval tasks, supporting context lengths of up to 8192 tokens. The model includes a set of task-specific Low-Rank Adaptation (LoRA) adapters to generate high-quality embeddings for query-document retrieval, clustering, classification, and text matching. Evaluation on the MTEB benchmark shows that jina-embeddings-v3 outperforms the latest proprietary embeddings from OpenAI and Cohere on English tasks, while achieving superior performance compared to multilingual-e5-large-instruct across all multilingual tasks. With a default output dimension of 1024, users can flexibly reduce the embedding dimensions to as low as 32 without compromising performance, enabled by Matryoshka Representation Learning.
A joint image reconstruction and segmentation approach based on disentangled representation learning was trained to enable cardiac cine MR imaging in real-time and under free-breathing. An exploratory feasibility study tested the proposed method in undersampled real-time acquisitions based on an in-house developed spiral bSSFP pulse sequence in eight healthy participants and five patients with intermittent atrial fibrillation. Images and predicted LV segmentations were compared to the reference standard of ECG-gated segmented Cartesian cine in repeated breath-holds and corresponding manual segmentation. On a 5-point Likert scale, image quality of the real-time breath-hold approach and Cartesian cine was comparable in healthy participants (RT-BH: 1.99 $\pm$ .98, Cartesian: 1.94 $\pm$ .86, p=.052), but slightly inferior in free-breathing (RT-FB: 2.40 $\pm$ .98, p<.001). In patients with arrhythmia, image quality from both real-time approaches was favourable (RT-BH: 2.10 $\pm$ 1.28, p<.001, RT-FB: 2.40 $\pm$ 1.13, p<.001, Cartesian: 2.68 $\pm$ 1.13). Intra-observer reliability was good (ICC=.77, 95%-confidence interval [.75, .79], p<.001). In functional analysis, a positive bias was observed for ejection fractions derived from the proposed model compared to the clinical reference standard (RT-BH mean EF: 58.5 $\pm$ 5.6%, bias: +3.47%, 95%-confidence interval [-.86, 7.79%], RT-FB mean: 57.9 $\pm$ 10.6%, bias: +1.45%, [-3.02, 5.91%], Cartesian mean: 54.9 $\pm$ 6.7%). The introduced real-time MR imaging technique is capable of acquiring high-quality cardiac cine data in 1-2 minutes without the need for ECG gating and breath-holds. It thus offers a promising alternative to the current clinical practice of segmented acquisition, with shorter scan times, higher patient comfort and increased robustness to arrhythmia and patient incompliance.
Graph Neural Networks (GNNs) have been widely employed for feature representation learning in molecular graphs. Therefore, it is crucial to enhance the expressiveness of feature representation to ensure the effectiveness of GNNs. However, a significant portion of current research primarily focuses on the structural features within individual molecules, often overlooking the structural similarity between molecules, which is a crucial aspect encapsulating rich information on the relationship between molecular properties and structural characteristics. Thus, these approaches fail to capture the rich semantic information at the molecular structure level. To bridge this gap, we introduce the \textbf{Molecular Structural Similarity Motif GNN (MSSM-GNN)}, a novel molecular graph representation learning method that can capture structural similarity information among molecules from a global perspective. In particular, we propose a specially designed graph that leverages graph kernel algorithms to represent the similarity between molecules quantitatively. Subsequently, we employ GNNs to learn feature representations from molecular graphs, aiming to enhance the accuracy of property prediction by incorporating additional molecular representation information. Finally, through a series of experiments conducted on both small-scale and large-scale molecular datasets, we demonstrate that our model consistently outperforms eleven state-of-the-art baselines. The codes are available at https://github.com/yaoyao-yaoyao-cell/MSSM-GNN.
Learning representations on large graphs is a long-standing challenge due to the inter-dependence nature. Transformers recently have shown promising performance on small graphs thanks to its global attention for capturing all-pair interactions beyond observed structures. Existing approaches tend to inherit the spirit of Transformers in language and vision tasks, and embrace complicated architectures by stacking deep attention-based propagation layers. In this paper, we attempt to evaluate the necessity of adopting multi-layer attentions in Transformers on graphs, which considerably restricts the efficiency. Specifically, we analyze a generic hybrid propagation layer, comprised of all-pair attention and graph-based propagation, and show that multi-layer propagation can be reduced to one-layer propagation, with the same capability for representation learning. It suggests a new technical path for building powerful and efficient Transformers on graphs, particularly through simplifying model architectures without sacrificing expressiveness. As exemplified by this work, we propose a Simplified Single-layer Graph Transformers (SGFormer), whose main component is a single-layer global attention that scales linearly w.r.t. graph sizes and requires none of any approximation for accommodating all-pair interactions. Empirically, SGFormer successfully scales to the web-scale graph ogbn-papers100M, yielding orders-of-magnitude inference acceleration over peer Transformers on medium-sized graphs, and demonstrates competitiveness with limited labeled data.
Multimodal language modeling has enabled breakthroughs for representation learning, yet remains unexplored in the realm of functional brain data for pathology detection. This paper pioneers EEG-language models (ELMs) trained on clinical reports and 15000 EEGs. We propose to combine multimodal alignment in this novel domain with timeseries cropping and text segmentation, enabling an extension based on multiple instance learning to alleviate misalignment between irrelevant EEG or text segments. Our multimodal models significantly improve pathology detection compared to EEG-only models across four evaluations and for the first time enable zero-shot classification as well as retrieval of both neural signals and reports. In sum, these results highlight the potential of ELMs, representing significant progress for clinical applications.
Self-supervised learning (SSL) has garnered significant attention in speech processing, excelling in linguistic tasks such as speech recognition. However, jointly improving the performance of pre-trained models on various downstream tasks, each requiring different speech information, poses significant challenges. To this purpose, we propose a progressive residual extraction based self-supervised learning method, named ProgRE. Specifically, we introduce two lightweight and specialized task modules into an encoder-style SSL backbone to enhance its ability to extract pitch variation and speaker information from speech. Furthermore, to prevent the interference of reinforced pitch variation and speaker information with irrelevant content information learning, we residually remove the information extracted by these two modules from the main branch. The main branch is then trained using HuBERT's speech masking prediction to ensure the performance of the Transformer's deep-layer features on content tasks. In this way, we can progressively extract pitch variation, speaker, and content representations from the input speech. Finally, we can combine multiple representations with diverse speech information using different layer weights to obtain task-specific representations for various downstream tasks. Experimental results indicate that our proposed method achieves joint performance improvements on various tasks, such as speaker identification, speech recognition, emotion recognition, speech enhancement, and voice conversion, compared to excellent SSL methods such as wav2vec2.0, HuBERT, and WavLM.
Climate change is a pressing global concern for governments, corporations, and citizens alike. This concern underscores the necessity for these entities to accurately assess the climate impact of manufacturing goods and providing services. Tools like process life cycle analysis (pLCA) are used to evaluate the climate impact of production, use, and disposal, from raw material mining through end-of-life. pLCA further enables practitioners to look deeply into material choices or manufacturing processes for individual parts, sub-assemblies, assemblies, and the final product. Reliable and detailed data on the life cycle stages and processes of the product or service under study are not always available or accessible, resulting in inaccurate assessment of climate impact. To overcome the data limitation and enhance the effectiveness of pLCA to generate an improved environmental impact profile, we are adopting an innovative strategy to identify alternative parts, products, and components that share similarities in terms of their form, function, and performance to serve as qualified substitutes. Focusing on enterprise electronics hardware, we propose a semi-supervised learning-based framework to identify substitute parts that leverages product bill of material (BOM) data and a small amount of component-level qualified substitute data (positive samples) to generate machine knowledge graph (MKG) and learn effective embeddings of the components that constitute electronic hardware. Our methodology is grounded in attributed graph embeddings and introduces a strategy to generate biased negative samples to significantly enhance the training process. We demonstrate improved performance and generalization over existing published models.
The landscape of social media content has evolved significantly, extending from text to multimodal formats. This evolution presents a significant challenge in combating misinformation. Previous research has primarily focused on single modalities or text-image combinations, leaving a gap in detecting multimodal misinformation. While the concept of entity consistency holds promise in detecting multimodal misinformation, simplifying the representation to a scalar value overlooks the inherent complexities of high-dimensional representations across different modalities. To address these limitations, we propose a Multimedia Misinformation Detection (MultiMD) framework for detecting misinformation from video content by leveraging cross-modal entity consistency. The proposed dual learning approach allows for not only enhancing misinformation detection performance but also improving representation learning of entity consistency across different modalities. Our results demonstrate that MultiMD outperforms state-of-the-art baseline models and underscore the importance of each modality in misinformation detection. Our research provides novel methodological and technical insights into multimodal misinformation detection.