Joint-embedding based learning (e.g., SimCLR, MoCo, DINO) and reconstruction-based learning (e.g., BEiT, SimMIM, MAE) are the two leading paradigms for self-supervised learning of vision transformers, but they differ substantially in their transfer performance. Here, we aim to explain these differences by analyzing the impact of these objectives on the structure and transferability of the learned representations. Our analysis reveals that reconstruction-based learning features are significantly dissimilar to joint-embedding based learning features and that models trained with similar objectives learn similar features even across architectures. These differences arise early in the network and are primarily driven by attention and normalization layers. We find that joint-embedding features yield better linear probe transfer for classification because the different objectives drive different distributions of information and invariances in the learned representation. These differences explain opposite trends in transfer performance for downstream tasks that require spatial specificity in features. Finally, we address how fine-tuning changes reconstructive representations to enable better transfer, showing that fine-tuning re-organizes the information to be more similar to pre-trained joint embedding models.
Self-supervised learning (SSL) techniques have recently been integrated into the few-shot learning (FSL) framework and have shown promising results in improving the few-shot image classification performance. However, existing SSL approaches used in FSL typically seek the supervision signals from the global embedding of every single image. Therefore, during the episodic training of FSL, these methods cannot capture and fully utilize the local visual information in image samples and the data structure information of the whole episode, which are beneficial to FSL. To this end, we propose to augment the few-shot learning objective with a novel self-supervised Episodic Spatial Pretext Task (ESPT). Specifically, for each few-shot episode, we generate its corresponding transformed episode by applying a random geometric transformation to all the images in it. Based on these, our ESPT objective is defined as maximizing the local spatial relationship consistency between the original episode and the transformed one. With this definition, the ESPT-augmented FSL objective promotes learning more transferable feature representations that capture the local spatial features of different images and their inter-relational structural information in each input episode, thus enabling the model to generalize better to new categories with only a few samples. Extensive experiments indicate that our ESPT method achieves new state-of-the-art performance for few-shot image classification on three mainstay benchmark datasets. The source code will be available at: https://github.com/Whut-YiRong/ESPT.
Human beings cooperatively navigate rule-constrained environments by adhering to mutually known navigational patterns, which may be represented as directional pathways or road lanes. Inferring these navigational patterns from incompletely observed environments is required for intelligent mobile robots operating in unmapped locations. However, algorithmically defining these navigational patterns is nontrivial. This paper presents the first self-supervised learning (SSL) method for learning to infer navigational patterns in real-world environments from partial observations only. We explain how geometric data augmentation, predictive world modeling, and an information-theoretic regularizer enables our model to predict an unbiased local directional soft lane probability (DSLP) field in the limit of infinite data. We demonstrate how to infer global navigational patterns by fitting a maximum likelihood graph to the DSLP field. Experiments show that our SSL model outperforms two SOTA supervised lane graph prediction models on the nuScenes dataset. We propose our SSL method as a scalable and interpretable continual learning paradigm for navigation by perception. Code is available at https://github.com/robin-karlsson0/dslp.
Understanding humans’ emotions is a challenge for computers. Nowadays, research on speech emotion recognition has been conducted progressively. Instead of a speech, affective information may lay on short vocal bursts (i.e., cry when sad). In this study, we evaluated a recent self-supervised learning model to extract acoustic embedding for affective vocal bursts tasks. There are four tasks investigated on both regression and classification problems. Using similar architectures, we found the effectiveness of using a pre-trained model over the baseline methods. The study is further expanded to evaluate the different number of seeds, patiences, and batch sizes on the performance of the four tasks.
Optical microscopy image plays an important role in scientific research through the direct visualization of the nanoworld, where the imaging mechanism is described as the convolution of the point spread function (PSF) and emitters. Based on a priori knowledge of the PSF or equivalent PSF, it is possible to achieve more precise exploration of the nanoworld. However, it is an outstanding challenge to directly extract the PSF from microscopy images. Here, with the help of self-supervised learning, we propose a physics-informed masked autoencoder (PiMAE) that enables a learnable estimation of the PSF and emitters directly from the raw microscopy images. We demonstrate our method in synthetic data and real-world experiments with significant accuracy and noise robustness. PiMAE outperforms DeepSTORM and the Richardson-Lucy algorithm in synthetic data tasks with an average improvement of 19.6\% and 50.7\% (35 tasks), respectively, as measured by the normalized root mean square error (NRMSE) metric. This is achieved without prior knowledge of the PSF, in contrast to the supervised approach used by DeepSTORM and the known PSF assumption in the Richardson-Lucy algorithm. Our method, PiMAE, provides a feasible scheme for achieving the hidden imaging mechanism in optical microscopy and has the potential to learn hidden mechanisms in many more systems.
We present view-synthesis autoencoders (VSA) in this paper, which is a self-supervised learning framework designed for vision transformers. Different from traditional 2D pretraining methods, VSA can be pre-trained with multi-view data. In each iteration, the input to VSA is one view (or multiple views) of a 3D object and the output is a synthesized image in another target pose. The decoder of VSA has several cross-attention blocks, which use the source view as value, source pose as key, and target pose as query. They achieve cross-attention to synthesize the target view. This simple approach realizes large-angle view synthesis and learns spatial invariant representation, where the latter is decent initialization for transformers on downstream tasks, such as 3D classification on ModelNet40, ShapeNet Core55, and ScanObjectNN. VSA outperforms existing methods significantly for linear probing and is competitive for fine-tuning. The code will be made publicly available.
This paper proposes a zero-shot text-to-speech (TTS) conditioned by a self-supervised speech-representation model acquired through self-supervised learning (SSL). Conventional methods with embedding vectors from x-vector or global style tokens still have a gap in reproducing the speaker characteristics of unseen speakers. A novel point of the proposed method is the direct use of the SSL model to obtain embedding vectors from speech representations trained with a large amount of data. We also introduce the separate conditioning of acoustic features and a phoneme duration predictor to obtain the disentangled embeddings between rhythm-based speaker characteristics and acoustic-feature-based ones. The disentangled embeddings will enable us to achieve better reproduction performance for unseen speakers and rhythm transfer conditioned by different speeches. Objective and subjective evaluations showed that the proposed method can synthesize speech with improved similarity and achieve speech-rhythm transfer.
The paper proposes a novel End-to-End Learning and Repair (E2ELR) architecture for training optimization proxies for economic dispatch problems. E2ELR combines deep neural networks with closed-form, differentiable repair layers, thereby integrating learning and feasibility in an end-to-end fashion. E2ELR is also trained with self-supervised learning, removing the need for labeled data and the solving of numerous optimization problems offline. E2ELR is evaluated on industry-size power grids with tens of thousands of buses using an economic dispatch that co-optimizes energy and reserves. The results demonstrate that the self-supervised E2ELR achieves state-of-the-art performance, with optimality gaps that outperform other baselines by at least an order of magnitude.
Self-supervised learning, dubbed the dark matter of intelligence, is a promising path to advance machine learning. Yet, much like cooking, training SSL methods is a delicate art with a high barrier to entry. While many components are familiar, successfully training a SSL method involves a dizzying set of choices from the pretext tasks to training hyper-parameters. Our goal is to lower the barrier to entry into SSL research by laying the foundations and latest SSL recipes in the style of a cookbook. We hope to empower the curious researcher to navigate the terrain of methods, understand the role of the various knobs, and gain the know-how required to explore how delicious SSL can be.
We introduce GEDI, a Bayesian framework that combines existing self-supervised learning objectives with likelihood-based generative models. This framework leverages the benefits of both GEnerative and DIscriminative approaches, resulting in improved symbolic representations over standalone solutions. Additionally, GEDI can be easily integrated and trained jointly with existing neuro-symbolic frameworks without the need for additional supervision or costly pre-training steps. We demonstrate through experiments on real-world data, including SVHN, CIFAR10, and CIFAR100, that GEDI outperforms existing self-supervised learning strategies in terms of clustering performance by a significant margin. The symbolic component further allows it to leverage knowledge in the form of logical constraints to improve performance in the small data regime.
Recent years have seen remarkable progress in speech emotion recognition (SER), thanks to advances in deep learning techniques. However, the limited availability of labeled data remains a significant challenge in the field. Self-supervised learning has recently emerged as a promising solution to address this challenge. In this paper, we propose the vector quantized masked autoencoder for speech (VQ-MAE-S), a self-supervised model that is fine-tuned to recognize emotions from speech signals. The VQ-MAE-S model is based on a masked autoencoder (MAE) that operates in the discrete latent space of a vector-quantized variational autoencoder. Experimental results show that the proposed VQ-MAE-S model, pre-trained on the VoxCeleb2 dataset and fine-tuned on emotional speech data, outperforms an MAE working on the raw spectrogram representation and other state-of-the-art methods in SER.
Deep metric learning aims to construct an embedding space where samples of the same class are close to each other, while samples of different classes are far away from each other. Most existing deep metric learning methods attempt to maximize the difference of inter-class features. And semantic related information is obtained by increasing the distance between samples of different classes in the embedding space. However, compressing all positive samples together while creating large margins between different classes unconsciously destroys the local structure between similar samples. Ignoring the intra-class variance contained in the local structure between similar samples, the embedding space obtained from training receives lower generalizability over unseen classes, which would lead to the network overfitting the training set and crashing on the test set. To address these considerations, this paper designs a self-supervised generative assisted ranking framework that provides a semi-supervised view of intra-class variance learning scheme for typical supervised deep metric learning. Specifically, this paper performs sample synthesis with different intensities and diversity for samples satisfying certain conditions to simulate the complex transformation of intra-class samples. And an intra-class ranking loss function is designed using the idea of self-supervised learning to constrain the network to maintain the intra-class distribution during the training process to capture the subtle intra-class variance. With this approach, a more realistic embedding space can be obtained in which global and local structures of samples are well preserved, thus enhancing the effectiveness of downstream tasks. Extensive experiments on four benchmarks have shown that this approach surpasses state-of-the-art methods
The normalized eight-point algorithm has been widely viewed as the cornerstone in two-view geometry computation, where the seminal Hartley's normalization has greatly improved the performance of the direct linear transformation algorithm. A natural question is, whether there exists and how to find other normalization methods that may further improve the performance as per each input sample. In this paper, we provide a novel perspective and propose two contributions to this fundamental problem: 1) we revisit the normalized eight-point algorithm and make a theoretical contribution by presenting the existence of different and better normalization algorithms; 2) we introduce a deep convolutional neural network with a self-supervised learning strategy for normalization. Given eight pairs of correspondences, our network directly predicts the normalization matrices, thus learning to normalize each input sample. Our learning-based normalization module can be integrated with both traditional (e.g., RANSAC) and deep learning frameworks (affording good interpretability) with minimal effort. Extensive experiments on both synthetic and real images demonstrate the effectiveness of our proposed approach.
With the development of the medical image field, researchers seek to develop a class of datasets to block the need for medical knowledge, such as \text{MedMNIST} (v2). MedMNIST (v2) includes a large number of small-sized (28 $\times$ 28 or 28 $\times$ 28 $\times$ 28) medical samples and the corresponding expert annotations (class label). The existing baseline model (Google AutoML Vision, ResNet-50+3D) can reach an average accuracy of over 70\% on MedMNIST (v2) datasets, which is comparable to the performance of expert decision-making. Nevertheless, we note that there are two insurmountable obstacles to modeling on MedMNIST (v2): 1) the raw images are cropped to low scales may cause effective recognition information to be dropped and the classifier to have difficulty in tracing accurate decision boundaries; 2) the labelers' subjective insight may cause many uncertainties in the label space. To address these issues, we develop a Complex Mixer (C-Mixer) with a pre-training framework to alleviate the problem of insufficient information and uncertainty in the label space by introducing an incentive imaginary matrix and a self-supervised scheme with random masking. Our method (incentive learning and self-supervised learning with masking) shows surprising potential on both the standard MedMNIST (v2) dataset, the customized weakly supervised datasets, and other image enhancement tasks.
Tremendous efforts have been made on document image rectification, but how to learn effective representation of such distorted images is still under-explored. In this paper, we present DocMAE, a novel self-supervised framework for document image rectification. Our motivation is to encode the structural cues in document images by leveraging masked autoencoder to benefit the rectification, i.e., the document boundaries, and text lines. Specifically, we first mask random patches of the background-excluded document images and then reconstruct the missing pixels. With such a self-supervised learning approach, the network is encouraged to learn the intrinsic structure of deformed documents by restoring document boundaries and missing text lines. Transfer performance in the downstream rectification task validates the effectiveness of our method. Extensive experiments are conducted to demonstrate the effectiveness of our method.
Few-shot relation reasoning on knowledge graphs (FS-KGR) aims to infer long-tail data-poor relations, which has drawn increasing attention these years due to its practicalities. The pre-training of previous methods needs to manually construct the meta-relation set, leading to numerous labor costs. Self-supervised learning (SSL) is treated as a solution to tackle the issue, but still at an early stage for FS-KGR task. Moreover, most of the existing methods ignore leveraging the beneficial information from aliasing relations (AR), i.e., data-rich relations with similar contextual semantics to the target data-poor relation. Therefore, we proposed a novel Self-Supervised Learning model by leveraging Aliasing Relations to assist FS-KGR, termed SARF. Concretely, four main components are designed in our model, i.e., SSL reasoning module, AR-assisted mechanism, fusion module, and scoring function. We first generate the representation of the co-occurrence patterns in a generative manner. Meanwhile, the representations of aliasing relations are learned to enhance reasoning in the AR-assist mechanism. Besides, multiple strategies, i.e., simple summation and learnable fusion, are offered for representation fusion. Finally, the generated representation is used for scoring. Extensive experiments on three few-shot benchmarks demonstrate that SARF achieves state-of-the-art performance compared with other methods in most cases.
The DCELANM-Net structure, which this article offers, is a model that ingeniously combines a Dual Channel Efficient Layer Aggregation Network (DCELAN) and a Micro Masked Autoencoder (Micro-MAE). On the one hand, for the DCELAN, the features are more effectively fitted by deepening the network structure; the deeper network can successfully learn and fuse the features, which can more accurately locate the local feature information; and the utilization of each layer of channels is more effectively improved by widening the network structure and residual connections. We adopted Micro-MAE as the learner of the model. In addition to being straightforward in its methodology, it also offers a self-supervised learning method, which has the benefit of being incredibly scaleable for the model.
We propose a self-supervised learning method for long text documents based on contrastive learning. A key to our method is Shuffle and Divide (SaD), a simple text augmentation algorithm that sets up a pretext task required for contrastive updates to BERT-based document embedding. SaD splits a document into two sub-documents containing randomly shuffled words in the entire documents. The sub-documents are considered positive examples, leaving all other documents in the corpus as negatives. After SaD, we repeat the contrastive update and clustering phases until convergence. It is naturally a time-consuming, cumbersome task to label text documents, and our method can help alleviate human efforts, which are most expensive resources in AI. We have empirically evaluated our method by performing unsupervised text classification on the 20 Newsgroups, Reuters-21578, BBC, and BBCSport datasets. In particular, our method pushes the current state-of-the-art, SS-SB-MT, on 20 Newsgroups by 20.94% in accuracy. We also achieve the state-of-the-art performance on Reuters-21578 and exceptionally-high accuracy performances (over 95%) for unsupervised classification on the BBC and BBCSport datasets.
The recent advances in representation learning inspire us to take on the challenging problem of unsupervised image classification tasks in a principled way. We propose ContraCluster, an unsupervised image classification method that combines clustering with the power of contrastive self-supervised learning. ContraCluster consists of three stages: (1) contrastive self-supervised pre-training (CPT), (2) contrastive prototype sampling (CPS), and (3) prototype-based semi-supervised fine-tuning (PB-SFT). CPS can select highly accurate, categorically prototypical images in an embedding space learned by contrastive learning. We use sampled prototypes as noisy labeled data to perform semi-supervised fine-tuning (PB-SFT), leveraging small prototypes and large unlabeled data to further enhance the accuracy. We demonstrate empirically that ContraCluster achieves new state-of-the-art results for standard benchmark datasets including CIFAR-10, STL-10, and ImageNet-10. For example, ContraCluster achieves about 90.8% accuracy for CIFAR-10, which outperforms DAC (52.2%), IIC (61.7%), and SCAN (87.6%) by a large margin. Without any labels, ContraCluster can achieve a 90.8% accuracy that is comparable to 95.8% by the best supervised counterpart.
Deep neural networks excel in supervised learning tasks but are constrained by the need for extensive labeled data. Self-supervised learning emerges as a promising alternative, allowing models to learn without explicit labels. Information theory, and notably the information bottleneck principle, has been pivotal in shaping deep neural networks. This principle focuses on optimizing the trade-off between compression and preserving relevant information, providing a foundation for efficient network design in supervised contexts. However, its precise role and adaptation in self-supervised learning remain unclear. In this work, we scrutinize various self-supervised learning approaches from an information-theoretic perspective, introducing a unified framework that encapsulates the \textit{self-supervised information-theoretic learning problem}. We weave together existing research into a cohesive narrative, delve into contemporary self-supervised methodologies, and spotlight potential research avenues and inherent challenges. Additionally, we discuss the empirical evaluation of information-theoretic quantities and their estimation methods. Overall, this paper furnishes an exhaustive review of the intersection of information theory, self-supervised learning, and deep neural networks.