Human perceives rich auditory experience with distinct sound heard by ears. Videos recorded with binaural audio particular simulate how human receives ambient sound. However, a large number of videos are with monaural audio only, which would degrade the user experience due to the lack of ambient information. To address this issue, we propose an audio spatialization framework to convert a monaural video into a binaural one exploiting the relationship across audio and visual components. By preserving the left-right consistency in both audio and visual modalities, our learning strategy can be viewed as a self-supervised learning technique, and alleviates the dependency on a large amount of video data with ground truth binaural audio data during training. Experiments on benchmark datasets confirm the effectiveness of our proposed framework in both semi-supervised and fully supervised scenarios, with ablation studies and visualization further support the use of our model for audio spatialization.
Bootstrap Your Own Latent (BYOL) introduced an approach to self-supervised learning avoiding the contrastive paradigm and subsequently removing the computational burden of negative sampling associated with such methods. However, we empirically find that the image representations produced under the BYOL's self-distillation paradigm are poorly distributed in representation space compared to contrastive methods. This work empirically demonstrates that feature diversity enforced by contrastive losses is beneficial to image representation uniformity when employed in BYOL, and as such, provides greater inter-class representation separability. Additionally, we explore and advocate the use of regularization methods, specifically the layer-wise minimization of hyperspherical energy (i.e. maximization of entropy) of network weights to encourage representation uniformity. We show that directly optimizing a measure of uniformity alongside the standard loss, or regularizing the networks of the BYOL architecture to minimize the hyperspherical energy of neurons can produce more uniformly distributed and therefore better performing representations for downstream tasks.
Transformer-based pretrained language models (PLMs) have started a new era in modern natural language processing (NLP). These models combine the power of transformers, transfer learning, and self-supervised learning (SSL). Following the success of these models in the general domain, the biomedical research community has developed various in-domain PLMs starting from BioBERT to the latest BioELECTRA and BioALBERT models. We strongly believe there is a need for a survey paper that can provide a comprehensive survey of various transformer-based biomedical pretrained language models (BPLMs). In this survey, we start with a brief overview of foundational concepts like self-supervised learning, embedding layer and transformer encoder layers. We discuss core concepts of transformer-based PLMs like pretraining methods, pretraining tasks, fine-tuning methods, and various embedding types specific to biomedical domain. We introduce a taxonomy for transformer-based BPLMs and then discuss all the models. We discuss various challenges and present possible solutions. We conclude by highlighting some of the open issues which will drive the research community to further improve transformer-based BPLMs.
In self-supervised representation learning, a common idea behind most of the state-of-the-art approaches is to enforce the robustness of the representations to predefined augmentations. A potential issue of this idea is the existence of completely collapsed solutions (i.e., constant features), which are typically avoided implicitly by carefully chosen implementation details. In this work, we study a relatively concise framework containing the most common components from recent approaches. We verify the existence of complete collapse and discover another reachable collapse pattern that is usually overlooked, namely dimensional collapse. We connect dimensional collapse with strong correlations between axes and consider such connection as a strong motivation for feature decorrelation (i.e., standardizing the covariance matrix). The gains from feature decorrelation are verified empirically to highlight the importance and the potential of this insight.
Self-supervised learning (SSL) has proven vital for advancing research in natural language processing (NLP) and computer vision (CV). The paradigm pretrains a shared model on large volumes of unlabeled data and achieves state-of-the-art (SOTA) for various tasks with minimal adaptation. However, the speech processing community lacks a similar setup to systematically explore the paradigm. To bridge this gap, we introduce Speech processing Universal PERformance Benchmark (SUPERB). SUPERB is a leaderboard to benchmark the performance of a shared model across a wide range of speech processing tasks with minimal architecture changes and labeled data. Among multiple usages of the shared model, we especially focus on extracting the representation learned from SSL due to its preferable re-usability. We present a simple framework to solve SUPERB tasks by learning task-specialized lightweight prediction heads on top of the frozen shared model. Our results demonstrate that the framework is promising as SSL representations show competitive generalizability and accessibility across SUPERB tasks. We release SUPERB as a challenge with a leaderboard and a benchmark toolkit to fuel the research in representation learning and general speech processing.
Self-supervised learning algorithms based on instance discrimination train encoders to be invariant to pre-defined transformations of the same instance. While most methods treat different views of the same image as positives for a contrastive loss, we are interested in using positives from other instances in the dataset. Our method, Nearest-Neighbor Contrastive Learning of visual Representations (NNCLR), samples the nearest neighbors from the dataset in the latent space, and treats them as positives. This provides more semantic variations than pre-defined transformations. We find that using the nearest-neighbor as positive in contrastive losses improves performance significantly on ImageNet classification, from 71.7% to 75.6%, outperforming previous state-of-the-art methods. On semi-supervised learning benchmarks we improve performance significantly when only 1% ImageNet labels are available, from 53.8% to 56.5%. On transfer learning benchmarks our method outperforms state-of-the-art methods (including supervised learning with ImageNet) on 8 out of 12 downstream datasets. Furthermore, we demonstrate empirically that our method is less reliant on complex data augmentations. We see a relative reduction of only 2.1% ImageNet Top-1 accuracy when we train using only random crops.
In this paper, we question if self-supervised learning provides new properties to Vision Transformer (ViT) that stand out compared to convolutional networks (convnets). Beyond the fact that adapting self-supervised methods to this architecture works particularly well, we make the following observations: first, self-supervised ViT features contain explicit information about the semantic segmentation of an image, which does not emerge as clearly with supervised ViTs, nor with convnets. Second, these features are also excellent k-NN classifiers, reaching 78.3% top-1 on ImageNet with a small ViT. Our study also underlines the importance of momentum encoder, multi-crop training, and the use of small patches with ViTs. We implement our findings into a simple self-supervised method, called DINO, which we interpret as a form of self-distillation with no labels. We show the synergy between DINO and ViTs by achieving 80.1% top-1 on ImageNet in linear evaluation with ViT-Base.
In this report, we relate the algorithmic design of Barlow Twins' method to the Hilbert-Schmidt Independence Criterion (HSIC), thus establishing it as a contrastive learning approach that is free of negative samples. Through this perspective, we argue that Barlow Twins (and thus the class of negative-sample-free contrastive learning methods) suggests a possibility to bridge the two major families of self-supervised learning philosophies: non-contrastive and contrastive approaches. In particular, Barlow twins exemplified how we could combine the best practices of both worlds: avoiding the need of large training batch size and negative sample pairing (like non-contrastive methods) and avoiding symmetry-breaking network designs (like contrastive methods).
Scenes play a crucial role in breaking the storyline of movies and TV episodes into semantically cohesive parts. However, given their complex temporal structure, finding scene boundaries can be a challenging task requiring large amounts of labeled training data. To address this challenge, we present a self-supervised shot contrastive learning approach (ShotCoL) to learn a shot representation that maximizes the similarity between nearby shots compared to randomly selected shots. We show how to apply our learned shot representation for the task of scene boundary detection to offer state-of-the-art performance on the MovieNet dataset while requiring only ~25% of the training labels, using 9x fewer model parameters and offering 7x faster runtime. To assess the effectiveness of ShotCoL on novel applications of scene boundary detection, we take on the problem of finding timestamps in movies and TV episodes where video-ads can be inserted while offering a minimally disruptive viewing experience. To this end, we collected a new dataset called AdCuepoints with 3,975 movies and TV episodes, 2.2 million shots and 19,119 minimally disruptive ad cue-point labels. We present a thorough empirical analysis on this dataset demonstrating the effectiveness of ShotCoL for ad cue-points detection.
Transfer learning has gained attention in medical image analysis due to limited annotated 3D medical datasets for training data-driven deep learning models in the real world. Existing 3D-based methods have transferred the pre-trained models to downstream tasks, which achieved promising results with only a small number of training samples. However, they demand a massive amount of parameters to train the model for 3D medical imaging. In this work, we propose a novel transfer learning framework, called Medical Transformer, that effectively models 3D volumetric images in the form of a sequence of 2D image slices. To make a high-level representation in 3D-form empowering spatial relations better, we take a multi-view approach that leverages plenty of information from the three planes of 3D volume, while providing parameter-efficient training. For building a source model generally applicable to various tasks, we pre-train the model in a self-supervised learning manner for masked encoding vector prediction as a proxy task, using a large-scale normal, healthy brain magnetic resonance imaging (MRI) dataset. Our pre-trained model is evaluated on three downstream tasks: (i) brain disease diagnosis, (ii) brain age prediction, and (iii) brain tumor segmentation, which are actively studied in brain MRI research. The experimental results show that our Medical Transformer outperforms the state-of-the-art transfer learning methods, efficiently reducing the number of parameters up to about 92% for classification and
Spatial reasoning on multi-view line drawings by state-of-the-art supervised deep networks is recently shown with puzzling low performances on the SPARE3D dataset. Based on the fact that self-supervised learning is helpful when a large number of data are available, we propose two self-supervised learning approaches to improve the baseline performance for view consistency reasoning and camera pose reasoning tasks on the SPARE3D dataset. For the first task, we use a self-supervised binary classification network to contrast the line drawing differences between various views of any two similar 3D objects, enabling the trained networks to effectively learn detail-sensitive yet view-invariant line drawing representations of 3D objects. For the second type of task, we propose a self-supervised multi-class classification framework to train a model to select the correct corresponding view from which a line drawing is rendered. Our method is even helpful for the downstream tasks with unseen camera poses. Experiments show that our method could significantly increase the baseline performance in SPARE3D, while some popular self-supervised learning methods cannot.
Data is the engine of modern computer vision, which necessitates collecting large-scale datasets. This is expensive, and guaranteeing the quality of the labels is a major challenge. In this paper, we investigate efficient annotation strategies for collecting multi-class classification labels for a large collection of images. While methods that exploit learnt models for labeling exist, a surprisingly prevalent approach is to query humans for a fixed number of labels per datum and aggregate them, which is expensive. Building on prior work on online joint probabilistic modeling of human annotations and machine-generated beliefs, we propose modifications and best practices aimed at minimizing human labeling effort. Specifically, we make use of advances in self-supervised learning, view annotation as a semi-supervised learning problem, identify and mitigate pitfalls and ablate several key design choices to propose effective guidelines for labeling. Our analysis is done in a more realistic simulation that involves querying human labelers, which uncovers issues with evaluation using existing worker simulation methods. Simulated experiments on a 125k image subset of the ImageNet100 show that it can be annotated to 80% top-1 accuracy with 0.35 annotations per image on average, a 2.7x and 6.7x improvement over prior work and manual annotation, respectively. Project page: https://fidler-lab.github.io/efficient-annotation-cookbook
Multimodal self-supervised learning is getting more and more attention as it allows not only to train large networks without human supervision but also to search and retrieve data across various modalities. In this context, this paper proposes a self-supervised training framework that learns a common multimodal embedding space that, in addition to sharing representations across different modalities, enforces a grouping of semantically similar instances. To this end, we extend the concept of instance-level contrastive learning with a multimodal clustering step in the training pipeline to capture semantic similarities across modalities. The resulting embedding space enables retrieval of samples across all modalities, even from unseen datasets and different domains. To evaluate our approach, we train our model on the HowTo100M dataset and evaluate its zero-shot retrieval capabilities in two challenging domains, namely text-to-video retrieval, and temporal action localization, showing state-of-the-art results on four different datasets.
Prior works on self-supervised pre-training focus on the joint training scenario, where massive unlabeled data are assumed to be given as input all at once, and only then is a learner trained. Unfortunately, such a problem setting is often impractical if not infeasible since many real-world tasks rely on sequential learning, e.g., data are decentralized or collected in a streaming fashion. In this paper, we conduct the first thorough and dedicated investigation on self-supervised pre-training with streaming data, aiming to shed light on the model behavior under this overlooked setup. Specifically, we pre-train over 500 models on four categories of pre-training streaming data from ImageNet and DomainNet and evaluate them on three types of downstream tasks and 12 different downstream datasets. Our studies show that, somehow beyond our expectation, with simple data replay or parameter regularization, sequential self-supervised pre-training turns out to be an efficient alternative for joint pre-training, as the performances of the former are mostly on par with those of the latter. Moreover, catastrophic forgetting, a common issue in sequential supervised learning, is much alleviated in sequential self-supervised learning (SSL), which is well justified through our comprehensive empirical analysis on representations and the sharpness of minima in the loss landscape. Our findings, therefore, suggest that, in practice, for SSL, the cumbersome joint training can be replaced mainly by sequential learning, which in turn enables a much broader spectrum of potential application scenarios.
Self-Supervised Learning (SSL) using huge unlabeled data has been successfully explored for image and natural language processing. Recent works also investigated SSL from speech. They were notably successful to improve performance on downstream tasks such as automatic speech recognition (ASR). While these works suggest it is possible to reduce dependence on labeled data for building efficient speech systems, their evaluation was mostly made on ASR and using multiple and heterogeneous experimental settings (most of them for English). This questions the objective comparison of SSL approaches and the evaluation of their impact on building speech systems. In this paper, we propose LeBenchmark: a reproducible framework for assessing SSL from speech. It not only includes ASR (high and low resource) tasks but also spoken language understanding, speech translation and emotion recognition. We also focus on speech technologies in a language different than English: French. SSL models of different sizes are trained from carefully sourced and documented datasets. Experiments show that SSL is beneficial for most but not all tasks which confirms the need for exhaustive and reliable benchmarks to evaluate its real impact. LeBenchmark is shared with the scientific community for reproducible research in SSL from speech.
Model Predictive Controllers (MPC) require a good model for the controlled process. In this paper I infer inductive biases about a physical system. I use these biases to derive a new neural network architecture that can model this real system that has noise and inertia. The main inductive biases exploited here are: the delayed impact of some inputs on the system and the separability between the temporal component and how the inputs interact to produce the output of a system. The inputs are independently delayed using shifted convolutional kernels. Feature interactions are modelled using a fully connected network that does not have access to temporal information. The available data and the problem setup allow the usage of Self Supervised Learning in order to train the models. The baseline architecture is an Attention based Reccurent network adapted to work with MPC like inputs. The proposed networks are faster, better at exploiting larger data volumes and are almost as good as baseline networks in terms of prediction performance. The proposed architecture family called Delay can be used in a real scenario to control systems with delayed responses with respect to its controls or inputs. Ablation studies show that the presence of delay kernels are vital to obtain any learning in proposed architecture. Code and some experimental data are available online.
We present a framework for learning multimodal representations from unlabeled data using convolution-free Transformer architectures. Specifically, our Video-Audio-Text Transformer (VATT) takes raw signals as inputs and extracts multimodal representations that are rich enough to benefit a variety of downstream tasks. We train VATT end-to-end from scratch using multimodal contrastive losses and evaluate its performance by the downstream tasks of video action recognition, audio event classification, image classification, and text-to-video retrieval. Furthermore, we study a modality-agnostic, single-backbone Transformer by sharing weights among the three modalities. We show that the convolution-free VATT outperforms state-of-the-art ConvNet-based architectures in the downstream tasks. Especially, VATT's vision Transformer achieves the top-1 accuracy of 82.1% on Kinetics-400, 83.6% on Kinetics-600, 72.7% on Kinetics-700, and 41.1% on Moments in Time, new records while avoiding supervised pre-training. Transferring to image classification leads to 78.7% top-1 accuracy on ImageNet compared to 64.7% by training the same Transformer from scratch, showing the generalizability of our model despite the domain gap between videos and images. VATT's audio Transformer also sets a new record on waveform-based audio event recognition by achieving the mAP of 39.4% on AudioSet without any supervised pre-training. VATT's source code is publicly available.
Learning from imprecise labels such as "animal" or "bird", but making precise predictions like "snow bunting" at inference time is an important capability for any classifier when expertly labeled training data is scarce. Contributions by volunteers or results of web crawling lack precision in this manner, but are still valuable. And crucially, these weakly labeled examples are available in larger quantities for lower cost than high-quality bespoke training data. CHILLAX, a recently proposed method to tackle this task, leverages a hierarchical classifier to learn from imprecise labels. However, it has two major limitations. First, it does not learn from examples labeled as the root of the hierarchy, e.g., "object". Second, an extrapolation of annotations to precise labels is only performed at test time, where confident extrapolations could be already used as training data. In this work, we extend CHILLAX with a self-supervised scheme using constrained semantic extrapolation to generate pseudo-labels. This addresses the second concern, which in turn solves the first problem, enabling an even weaker supervision requirement than CHILLAX. We evaluate our approach empirically, showing that our method allows for a consistent accuracy improvement of 0.84 to 1.19 percent points over CHILLAX and is suitable as a drop-in replacement without any negative consequences such as longer training times.
In medical image analysis, the cost of acquiring high-quality data and their annotation by experts is a barrier in many medical applications. Most of the techniques used are based on supervised learning framework and need a large amount of annotated data to achieve satisfactory performance. As an alternative, in this paper, we propose a self-supervised learning (SSL) approach to learn the spatial anatomical representations from the frames of magnetic resonance (MR) video clips for the diagnosis of knee medical conditions. The pretext model learns meaningful spatial context-invariant representations. The downstream task in our paper is a class imbalanced multi-label classification. Different experiments show that the features learnt by the pretext model provide competitive performance in the downstream task. Moreover, the efficiency and reliability of the proposed pretext model in learning representations of minority classes without applying any strategy towards imbalance in the dataset can be seen from the results. To the best of our knowledge, this work is the first work of its kind in showing the effectiveness and reliability of self-supervised learning algorithms in class imbalanced multi-label classification tasks on MR videos. The code for evaluation of the proposed work is available at https://github.com/sadimanna/skid.
Text encodings from automatic speech recognition (ASR) transcripts and audio representations have shown promise in speech emotion recognition (SER) ever since. Yet, it is challenging to explain the effect of each information stream on the SER systems. Further, more clarification is required for analysing the impact of ASR's word error rate (WER) on linguistic emotion recognition per se and in the context of fusion with acoustic information exploitation in the age of deep ASR systems. In order to tackle the above issues, we create transcripts from the original speech by applying three modern ASR systems, including an end-to-end model trained with recurrent neural network-transducer loss, a model with connectionist temporal classification loss, and a wav2vec framework for self-supervised learning. Afterwards, we use pre-trained textual models to extract text representations from the ASR outputs and the gold standard. For extraction and learning of acoustic speech features, we utilise openSMILE, openXBoW, DeepSpectrum, and auDeep. Finally, we conduct decision-level fusion on both information streams -- acoustics and linguistics. Using the best development configuration, we achieve state-of-the-art unweighted average recall values of $73.6\,\%$ and $73.8\,\%$ on the speaker-independent development and test partitions of IEMOCAP, respectively.