Reinforcement learning can solve decision-making problems and train an agent to behave in an environment according to a predesigned reward function. However, such an approach becomes very problematic if the reward is too sparse and so the agent does not come across the reward during the environmental exploration. The solution to such a problem may be to equip the agent with an intrinsic motivation that will provide informed exploration during which the agent is likely to also encounter external reward. Novelty detection is one of the promising branches of intrinsic motivation research. We present Self-supervised Network Distillation (SND), a class of intrinsic motivation algorithms based on the distillation error as a novelty indicator, where the predictor model and the target model are both trained. We adapted three existing self-supervised methods for this purpose and experimentally tested them on a set of ten environments that are considered difficult to explore. The results show that our approach achieves faster growth and higher external reward for the same training time compared to the baseline models, which implies improved exploration in a very sparse reward environment. In addition, the analytical methods we applied provide valuable explanatory insights into our proposed models.
Backdoor data detection is traditionally studied in an end-to-end supervised learning (SL) setting. However, recent years have seen the proliferating adoption of self-supervised learning (SSL) and transfer learning (TL), due to their lesser need for labeled data. Successful backdoor attacks have also been demonstrated in these new settings. However, we lack a thorough understanding of the applicability of existing detection methods across a variety of learning settings. By evaluating 56 attack settings, we show that the performance of most existing detection methods varies significantly across different attacks and poison ratios, and all fail on the state-of-the-art clean-label attack. In addition, they either become inapplicable or suffer large performance losses when applied to SSL and TL. We propose a new detection method called Active Separation via Offset (ASSET), which actively induces different model behaviors between the backdoor and clean samples to promote their separation. We also provide procedures to adaptively select the number of suspicious points to remove. In the end-to-end SL setting, ASSET is superior to existing methods in terms of consistency of defensive performance across different attacks and robustness to changes in poison ratios; in particular, it is the only method that can detect the state-of-the-art clean-label attack. Moreover, ASSET's average detection rates are higher than the best existing methods in SSL and TL, respectively, by 69.3% and 33.2%, thus providing the first practical backdoor defense for these new DL settings. We open-source the project to drive further development and encourage engagement: https://github.com/ruoxi-jia-group/ASSET.
The online emergence of multi-modal sharing platforms (eg, TikTok, Youtube) is powering personalized recommender systems to incorporate various modalities (eg, visual, textual and acoustic) into the latent user representations. While existing works on multi-modal recommendation exploit multimedia content features in enhancing item embeddings, their model representation capability is limited by heavy label reliance and weak robustness on sparse user behavior data. Inspired by the recent progress of self-supervised learning in alleviating label scarcity issue, we explore deriving self-supervision signals with effectively learning of modality-aware user preference and cross-modal dependencies. To this end, we propose a new Multi-Modal Self-Supervised Learning (MMSSL) method which tackles two key challenges. Specifically, to characterize the inter-dependency between the user-item collaborative view and item multi-modal semantic view, we design a modality-aware interactive structure learning paradigm via adversarial perturbations for data augmentation. In addition, to capture the effects that user's modality-aware interaction pattern would interweave with each other, a cross-modal contrastive learning approach is introduced to jointly preserve the inter-modal semantic commonality and user preference diversity. Experiments on real-world datasets verify the superiority of our method in offering great potential for multimedia recommendation over various state-of-the-art baselines. The implementation is released at: https://github.com/HKUDS/MMSSL.
Recent progress has been made towards learning invariant or equivariant representations with self-supervised learning. While invariant methods are evaluated on large scale datasets, equivariant ones are evaluated in smaller, more controlled, settings. We aim at bridging the gap between the two in order to learn more diverse representations that are suitable for a wide range of tasks. We start by introducing a dataset called 3DIEBench, consisting of renderings from 3D models over 55 classes and more than 2.5 million images where we have full control on the transformations applied to the objects. We further introduce a predictor architecture based on hypernetworks to learn equivariant representations with no possible collapse to invariance. We introduce SIE (Split Invariant-Equivariant) which combines the hypernetwork-based predictor with representations split in two parts, one invariant, the other equivariant, to learn richer representations. We demonstrate significant performance gains over existing methods on equivariance related tasks from both a qualitative and quantitative point of view. We further analyze our introduced predictor and show how it steers the learned latent space. We hope that both our introduced dataset and approach will enable learning richer representations without supervision in more complex scenarios. Code and data are available at https://github.com/facebookresearch/SIE.
Fully supervised salient object detection (SOD) methods have made considerable progress in performance, yet these models rely heavily on expensive pixel-wise labels. Recently, to achieve a trade-off between labeling burden and performance, scribble-based SOD methods have attracted increasing attention. Previous scribble-based models directly implement the SOD task only based on SOD training data with limited information, it is extremely difficult for them to understand the image and further achieve a superior SOD task. In this paper, we propose a simple yet effective framework guided by general visual representations with rich contextual semantic knowledge for scribble-based SOD. These general visual representations are generated by self-supervised learning based on large-scale unlabeled datasets. Our framework consists of a task-related encoder, a general visual module, and an information integration module to efficiently combine the general visual representations with task-related features to perform the SOD task based on understanding the contextual connections of images. Meanwhile, we propose a novel global semantic affinity loss to guide the model to perceive the global structure of the salient objects. Experimental results on five public benchmark datasets demonstrate that our method, which only utilizes scribble annotations without introducing any extra label, outperforms the state-of-the-art weakly supervised SOD methods. Specifically, it outperforms the previous best scribble-based method on all datasets with an average gain of 5.5% for max f-measure, 5.8% for mean f-measure, 24% for MAE, and 3.1% for E-measure. Moreover, our method achieves comparable or even superior performance to the state-of-the-art fully supervised models.
Large-scale volumetric medical images with annotation are rare, costly, and time prohibitive to acquire. Self-supervised learning (SSL) offers a promising pre-training and feature extraction solution for many downstream tasks, as it only uses unlabeled data. Recently, SSL methods based on instance discrimination have gained popularity in the medical imaging domain. However, SSL pre-trained encoders may use many clues in the image to discriminate an instance that are not necessarily disease-related. Moreover, pathological patterns are often subtle and heterogeneous, requiring the ability of the desired method to represent anatomy-specific features that are sensitive to abnormal changes in different body parts. In this work, we present a novel SSL framework, named DrasCLR, for 3D medical imaging to overcome these challenges. We propose two domain-specific contrastive learning strategies: one aims to capture subtle disease patterns inside a local anatomical region, and the other aims to represent severe disease patterns that span larger regions. We formulate the encoder using conditional hyper-parameterized network, in which the parameters are dependant on the anatomical location, to extract anatomically sensitive features. Extensive experiments on large-scale computer tomography (CT) datasets of lung images show that our method improves the performance of many downstream prediction and segmentation tasks. The patient-level representation improves the performance of the patient survival prediction task. We show how our method can detect emphysema subtypes via dense prediction. We demonstrate that fine-tuning the pre-trained model can significantly reduce annotation efforts without sacrificing emphysema detection accuracy. Our ablation study highlights the importance of incorporating anatomical context into the SSL framework.
Costly, noisy, and over-specialized, labels are to be set aside in favor of unsupervised learning if we hope to learn cheap, reliable, and transferable models. To that end, spectral embedding, self-supervised learning, or generative modeling have offered competitive solutions. Those methods however come with numerous challenges \textit{e.g.} estimating geodesic distances, specifying projector architectures and anti-collapse losses, or specifying decoder architectures and reconstruction losses. In contrast, we introduce a simple explainable alternative -- coined \textbf{DIET} -- to learn representations from unlabeled data, free of those challenges. \textbf{DIET} is blatantly simple: take one's favorite classification setup and use the \textbf{D}atum \textbf{I}nd\textbf{E}x as its \textbf{T}arget class, \textit{i.e. each sample is its own class}, no further changes needed. \textbf{DIET} works without a decoder/projector network, is not based on positive pairs nor reconstruction, introduces no hyper-parameters, and works out-of-the-box across datasets and architectures. Despite \textbf{DIET}'s simplicity, the learned representations are of high-quality and often on-par with the state-of-the-art \textit{e.g.} using a linear classifier on top of DIET's learned representation reaches $71.4\%$ on CIFAR100 with a Resnet101, $52.5\%$ on TinyImagenet with a Resnext50.
An established normative approach for understanding the algorithmic basis of neural computation is to derive online algorithms from principled computational objectives and evaluate their compatibility with anatomical and physiological observations. Similarity matching objectives have served as successful starting points for deriving online algorithms that map onto neural networks (NNs) with point neurons and Hebbian/anti-Hebbian plasticity. These NN models account for many anatomical and physiological observations; however, the objectives have limited computational power and the derived NNs do not explain multi-compartmental neuronal structures and non-Hebbian forms of plasticity that are prevalent throughout the brain. In this article, we unify and generalize recent extensions of the similarity matching approach to address more complex objectives, including a large class of unsupervised and self-supervised learning tasks that can be formulated as symmetric generalized eigenvalue problems or nonnegative matrix factorization problems. Interestingly, the online algorithms derived from these objectives naturally map onto NNs with multi-compartmental neurons and local, non-Hebbian learning rules. Therefore, this unified extension of the similarity matching approach provides a normative framework that facilitates understanding multi-compartmental neuronal structures and non-Hebbian plasticity found throughout the brain.
A typical fluency scoring system generally relies on an automatic speech recognition (ASR) system to obtain time stamps in input speech for either the subsequent calculation of fluency-related features or directly modeling speech fluency with an end-to-end approach. This paper describes a novel ASR-free approach for automatic fluency assessment using self-supervised learning (SSL). Specifically, wav2vec2.0 is used to extract frame-level speech features, followed by K-means clustering to assign a pseudo label (cluster index) to each frame. A BLSTM-based model is trained to predict an utterance-level fluency score from frame-level SSL features and the corresponding cluster indexes. Neither speech transcription nor time stamp information is required in the proposed system. It is ASR-free and can potentially avoid the ASR errors effect in practice. Experimental results carried out on non-native English databases show that the proposed approach significantly improves the performance in the "open response" scenario as compared to previous methods and matches the recently reported performance in the "read aloud" scenario.
Recent years have witnessed a boom in self-supervised learning (SSL) in various areas including speech processing. Speech based SSL models present promising performance in a range of speech related tasks. However, the training of SSL models is computationally expensive and a common practice is to fine-tune a released SSL model on the specific task. It is essential to use consistent front-end input during pre-training and fine-tuning. This consistency may introduce potential issues when the optimal front-end is not the same as that used in pre-training. In this paper, we propose a simple but effective front-end adapter to address this front-end discrepancy. By minimizing the distance between the outputs of different front-ends, the filterbank feature (Fbank) can be compatible with SSL models which are pre-trained with waveform. The experiment results demonstrate the effectiveness of our proposed front-end adapter on several popular SSL models for the speech recognition task.
State estimation is one of the greatest challenges for cloth manipulation due to cloth's high dimensionality and self-occlusion. Prior works propose to identify the full state of crumpled clothes by training a mesh reconstruction model in simulation. However, such models are prone to suffer from a sim-to-real gap due to differences between cloth simulation and the real world. In this work, we propose a self-supervised method to finetune a mesh reconstruction model in the real world. Since the full mesh of crumpled cloth is difficult to obtain in the real world, we design a special data collection scheme and an action-conditioned model-based cloth tracking method to generate pseudo-labels for self-supervised learning. By finetuning the pretrained mesh reconstruction model on this pseudo-labeled dataset, we show that we can improve the quality of the reconstructed mesh without requiring human annotations, and improve the performance of downstream manipulation task.
While enabling accelerated acquisition and improved reconstruction accuracy, current deep MRI reconstruction networks are typically supervised, require fully sampled data, and are limited to Cartesian sampling patterns. These factors limit their practical adoption as fully-sampled MRI is prohibitively time-consuming to acquire clinically. Further, non-Cartesian sampling patterns are particularly desirable as they are more amenable to acceleration and show improved motion robustness. To this end, we present a fully self-supervised approach for accelerated non-Cartesian MRI reconstruction which leverages self-supervision in both k-space and image domains. In training, the undersampled data are split into disjoint k-space domain partitions. For the k-space self-supervision, we train a network to reconstruct the input undersampled data from both the disjoint partitions and from itself. For the image-level self-supervision, we enforce appearance consistency obtained from the original undersampled data and the two partitions. Experimental results on our simulated multi-coil non-Cartesian MRI dataset demonstrate that DDSS can generate high-quality reconstruction that approaches the accuracy of the fully supervised reconstruction, outperforming previous baseline methods. Finally, DDSS is shown to scale to highly challenging real-world clinical MRI reconstruction acquired on a portable low-field (0.064 T) MRI scanner with no data available for supervised training while demonstrating improved image quality as compared to traditional reconstruction, as determined by a radiologist study.
Self-supervised learning (SSL) learns high-quality representations from large pools of unlabeled training data. As datasets grow larger, it becomes crucial to identify the examples that contribute the most to learning such representations. This enables efficient SSL by reducing the volume of data required. Nevertheless, quantifying the value of examples for SSL has remained an open question. In this work, we address this problem for the first time, by proving that examples that contribute the most to contrastive SSL are those that have the most similar augmentations to other examples, in expectation. We provide rigorous guarantees for the generalization performance of contrastive learning on such subsets. Through extensive experiments, we show that we can safely exclude 20% of examples from CIFAR100 and 40% from STL10 and TinyImageNet, without affecting downstream task performance. In general, subsets selected by our method outperform random subsets by over 3% across these datasets. Interestingly, we also discover the subsets that contribute the most to contrastive learning are those that contribute the least to supervised learning. Code available at https://github.com/bigml-cs-ucla/sas-data-efficient-contrastive-learning.
The computer-aided disease diagnosis from radiomic data is important in many medical applications. However, developing such a technique relies on annotating radiological images, which is a time-consuming, labor-intensive, and expensive process. In this work, we present the first novel collaborative self-supervised learning method to solve the challenge of insufficient labeled radiomic data, whose characteristics are different from text and image data. To achieve this, we present two collaborative pretext tasks that explore the latent pathological or biological relationships between regions of interest and the similarity and dissimilarity information between subjects. Our method collaboratively learns the robust latent feature representations from radiomic data in a self-supervised manner to reduce human annotation efforts, which benefits the disease diagnosis. We compared our proposed method with other state-of-the-art self-supervised learning methods on a simulation study and two independent datasets. Extensive experimental results demonstrated that our method outperforms other self-supervised learning methods on both classification and regression tasks. With further refinement, our method shows the potential advantage in automatic disease diagnosis with large-scale unlabeled data available.
Self-supervised feature learning enables perception systems to benefit from the vast raw data recorded by vehicle fleets worldwide. While video-level self-supervised learning approaches have shown strong generalizability on classification tasks, the potential to learn dense representations from sequential data has been relatively unexplored. In this work, we propose TempO, a temporal ordering pretext task for pre-training region-level feature representations for perception tasks. We embed each frame by an unordered set of proposal feature vectors, a representation that is natural for object detection or tracking systems, and formulate the sequential ordering by predicting frame transition probabilities in a transformer-based multi-frame architecture whose complexity scales less than quadratic with respect to the sequence length. Extensive evaluations on the BDD100K, nuImages, and MOT17 datasets show that our TempO pre-training approach outperforms single-frame self-supervised learning methods as well as supervised transfer learning initialization strategies, achieving an improvement of +0.7% in mAP for object detection and +2.0% in the HOTA score for multi-object tracking.
Self-supervised learning has demonstrated remarkable capability in representation learning for skeleton-based action recognition. Existing methods mainly focus on applying global data augmentation to generate different views of the skeleton sequence for contrastive learning. However, due to the rich action clues in the skeleton sequences, existing methods may only take a global perspective to learn to discriminate different skeletons without thoroughly leveraging the local relationship between different skeleton joints and video frames, which is essential for real-world applications. In this work, we propose a Partial Spatio-Temporal Learning (PSTL) framework to exploit the local relationship from a partial skeleton sequences built by a unique spatio-temporal masking strategy. Specifically, we construct a negative-sample-free triplet steam structure that is composed of an anchor stream without any masking, a spatial masking stream with Central Spatial Masking (CSM), and a temporal masking stream with Motion Attention Temporal Masking (MATM). The feature cross-correlation matrix is measured between the anchor stream and the other two masking streams, respectively. (1) Central Spatial Masking discards selected joints from the feature calculation process, where the joints with a higher degree of centrality have a higher possibility of being selected. (2) Motion Attention Temporal Masking leverages the motion of action and remove frames that move faster with a higher possibility. Our method achieves SOTA performance on NTU-60, NTU-120 and PKU-MMD under various downstream tasks. A practical evaluation is performed where some skeleton joints are lost in downstream tasks. In contrast to previous methods that suffer from large performance drops, our PSTL can still achieve remarkable results, validating the robustness of our method. Code: https://github.com/YujieOuO/PSTL.git.
In this work, we propose a zero-shot voice conversion method using speech representations trained with self-supervised learning. First, we develop a multi-task model to decompose a speech utterance into features such as linguistic content, speaker characteristics, and speaking style. To disentangle content and speaker representations, we propose a training strategy based on Siamese networks that encourages similarity between the content representations of the original and pitch-shifted audio. Next, we develop a synthesis model with pitch and duration predictors that can effectively reconstruct the speech signal from its decomposed representation. Our framework allows controllable and speaker-adaptive synthesis to perform zero-shot any-to-any voice conversion achieving state-of-the-art results on metrics evaluating speaker similarity, intelligibility, and naturalness. Using just 10 seconds of data for a target speaker, our framework can perform voice swapping and achieves a speaker verification EER of 5.5% for seen speakers and 8.4% for unseen speakers.
We propose a self-supervised learning approach for videos that learns representations of both the RGB frames and the accompanying audio without human supervision. In contrast to images that capture the static scene appearance, videos also contain sound and temporal scene dynamics. To leverage the temporal and aural dimension inherent to videos, our method extends temporal self-supervision to the audio-visual setting and integrates it with multi-modal contrastive objectives. As temporal self-supervision, we pose playback speed and direction recognition in both modalities and propose intra- and inter-modal temporal ordering tasks. Furthermore, we design a novel contrastive objective in which the usual pairs are supplemented with additional sample-dependent positives and negatives sampled from the evolving feature space. In our model, we apply such losses among video clips and between videos and their temporally corresponding audio clips. We verify our model design in extensive ablation experiments and evaluate the video and audio representations in transfer experiments to action recognition and retrieval on UCF101 and HMBD51, audio classification on ESC50, and robust video fingerprinting on VGG-Sound, with state-of-the-art results.
Blazars are active galactic nuclei with relativistic jets pointed almost directly at Earth. Blazars are characterized by strong, apparently stochastic flux variability at virtually all observed wavelengths and timescales, from minutes to years, the physical origin of which is still poorly understood. In the high-energy gamma-ray band, the Large Area Telescope aboard the Fermi space telescope (Fermi-LAT) has conducted regular monitoring of thousands of blazars since 2008. Deep learning can help uncover structure in gamma-ray blazars' complex variability patterns that traditional methods based on parametric statistical modeling or manual feature engineering may miss. In this work, we propose using a self-supervised Transformer encoder architecture to construct an effective representation of blazar gamma-ray variability. Measurement errors, upper limits, and missing data are accommodated using learned encodings. The model predicts a set of quantiles for the flux probability distribution at each time step, an architecture naturally suited for describing data generated by a stochastic process. As a proof of concept for how the model output can be analyzed to extract scientifically relevant information, a preliminary search for weekly-timescale time-reversal asymmetry in gamma-ray blazar light curves was conducted, finding no significant evidence for asymmetry.
This paper describes recent developments in object specific pose and shape prediction from single images. The main contribution is a new approach to camera pose prediction by self-supervised learning of keypoints corresponding to locations on a category specific deformable shape. We designed a network to generate a proxy ground-truth heatmap from a set of keypoints distributed all over the category-specific mean shape, where each is represented by a unique color on a labeled texture. The proxy ground-truth heatmap is used to train a deep keypoint prediction network, which can be used in online inference. The proposed approach to camera pose prediction show significant improvements when compared with state-of-the-art methods. Our approach to camera pose prediction is used to infer 3D objects from 2D image frames of video sequences online. To train the reconstruction model, it receives only a silhouette mask from a single frame of a video sequence in every training step and a category-specific mean object shape. We conducted experiments using three different datasets representing the bird category: the CUB [51] image dataset, YouTubeVos and the Davis video datasets. The network is trained on the CUB dataset and tested on all three datasets. The online experiments are demonstrated on YouTubeVos and Davis [56] video sequences using a network trained on the CUB training set.