In regularization Self-Supervised Learning (SSL) methods for graphs, computational complexity increases with the number of nodes in graphs and embedding dimensions. To mitigate the scalability of non-contrastive graph SSL, we propose a novel approach to reduce the cost of computing the covariance matrix for the pre-training loss function with volume-maximization terms. Our work focuses on reducing the cost associated with the loss computation via graph node or dimension sampling. We provide theoretical insight into why dimension sampling would result in accurate loss computations and support it with mathematical derivation of the novel approach. We develop our experimental setup on the node-level graph prediction tasks, where SSL pre-training has shown to be difficult due to the large size of real world graphs. Our experiments demonstrate that the cost associated with the loss computation can be reduced via node or dimension sampling without lowering the downstream performance. Our results demonstrate that sampling mostly results in improved downstream performance. Ablation studies and experimental analysis are provided to untangle the role of the different factors in the experimental setup.
Consider a predictor, a learner, whose input is a stream of discrete items. The predictor's task, at every time point, is probabilistic multiclass prediction, i.e. to predict which item may occur next by outputting zero or more candidate items, each with a probability, after which the actual item is revealed and the predictor updates. To output probabilities, the predictor keeps track of the proportions of the items it has seen. The stream is unbounded (lifelong), and the predictor has finite limited space. The task is open-ended: the set of items is unknown to the predictor and their totality can also grow unbounded. Moreover, there is non-stationarity: the underlying frequencies of items may change, substantially, from time to time. For instance, new items may start appearing and a few recently frequent items may cease to occur again. The predictor, being space-bounded, need only provide probabilities for those items which, at the time of prediction, have sufficiently high frequency, i.e., the salient items. This problem is motivated in the setting of Prediction Games, a self-supervised learning regime where concepts serve as both the predictors and the predictands, and the set of concepts grows over time, resulting in non-stationarities as new concepts are generated and used. We design and study a number of predictors, sparse moving averages(SMAs), for the task. One SMA adapts the sparse exponentiated moving average and another is based on queuing a few counts, keeping dynamic per-item histories. Evaluating the predicted probabilities, under noise and non-stationarity, presents challenges, and we discuss and develop evaluation methods, one based on bounding log-loss. We show that a combination of ideas, supporting dynamic predictand-specific learning rates, offers advantages in terms of faster adaption to change (plasticity), while also supporting low variance (stability).
Repetitive DNA (repeats) poses significant challenges for accurate and efficient genome assembly and sequence alignment. This is particularly true for metagenomic data, where genome dynamics such as horizontal gene transfer, gene duplication, and gene loss/gain complicate accurate genome assembly from metagenomic communities. Detecting repeats is a crucial first step in overcoming these challenges. To address this issue, we propose GraSSRep, a novel approach that leverages the assembly graph's structure through graph neural networks (GNNs) within a self-supervised learning framework to classify DNA sequences into repetitive and non-repetitive categories. Specifically, we frame this problem as a node classification task within a metagenomic assembly graph. In a self-supervised fashion, we rely on a high-precision (but low-recall) heuristic to generate pseudo-labels for a small proportion of the nodes. We then use those pseudo-labels to train a GNN embedding and a random forest classifier to propagate the labels to the remaining nodes. In this way, GraSSRep combines sequencing features with pre-defined and learned graph features to achieve state-of-the-art performance in repeat detection. We evaluate our method using simulated and synthetic metagenomic datasets. The results on the simulated data highlight our GraSSRep's robustness to repeat attributes, demonstrating its effectiveness in handling the complexity of repeated sequences. Additionally, our experiments with synthetic metagenomic datasets reveal that incorporating the graph structure and the GNN enhances our detection performance. Finally, in comparative analyses, GraSSRep outperforms existing repeat detection tools with respect to precision and recall.
The standard approach to modern self-supervised learning is to generate random views through data augmentations and minimise a loss computed from the representations of these views. This inherently encourages invariance to the transformations that comprise the data augmentation function. In this work, we show that adding a module to constrain the representations to be predictive of an affine transformation improves the performance and efficiency of the learning process. The module is agnostic to the base self-supervised model and manifests in the form of an additional loss term that encourages an aggregation of the encoder representations to be predictive of an affine transformation applied to the input images. We perform experiments in various modern self-supervised models and see a performance improvement in all cases. Further, we perform an ablation study on the components of the affine transformation to understand which of them is affecting performance the most, as well as on key architectural design decisions.
The increasing variety and quantity of tagged multimedia content on a variety of online platforms offer a unique opportunity to advance the field of human action recognition. In this study, we utilize 283,582 unique, unlabeled TikTok video clips, categorized into 386 hashtags, to train a domain-specific foundation model for action recognition. We employ VideoMAE V2, an advanced model integrating Masked Autoencoders (MAE) with Vision Transformers (ViT), pre-trained on this diverse collection of unstructured videos. Our model, fine-tuned on established action recognition benchmarks such as UCF101 and HMDB51, achieves state-of-the-art results: 99.05% on UCF101, 86.08% on HMDB51, 85.51% on Kinetics-400, and 74.27% on Something-Something V2 using the ViT-giant backbone. These results highlight the potential of using unstructured and unlabeled videos as a valuable source of diverse and dynamic content for training foundation models. Our investigation confirms that while initial increases in pre-training data volume significantly enhance model performance, the gains diminish as the dataset size continues to expand. Our findings emphasize two critical axioms in self-supervised learning for computer vision: (1) additional pre-training data can yield diminishing benefits for some datasets and (2) quality is more important than quantity in self-supervised learning, especially when building foundation models.
When LLMs perform zero-shot inference, they typically use a prompt with a task specification, and generate a completion. However, there is no work to explore the possibility of the reverse - going from completion to task specification. In this paper, we employ both directions to perform cycle-supervised learning entirely in-context. Our goal is to create a forward map f : X -> Y (e.g. image -> generated caption), coupled with a backward map g : Y -> X (e.g. caption -> generated image) to construct a cycle-consistency "loss" (formulated as an update to the prompt) to enforce g(f(X)) ~= X. The technique, called CyclePrompt, uses cycle-consistency as a free supervisory signal to iteratively craft the prompt. Importantly, CyclePrompt reinforces model performance without expensive fine-tuning, without training data, and without the complexity of external environments (e.g. compilers, APIs). We demonstrate CyclePrompt in two domains: code generation and image captioning. Our results on the HumanEval coding benchmark put us in first place on the leaderboard among models that do not rely on extra training data or usage of external environments, and third overall. Compared to the GPT4 baseline, we improve accuracy from 80.5% to 87.2%. In the vision-language space, we generate detailed image captions which outperform baseline zero-shot GPT4V captions, when tested against natural (VQAv2) and diagrammatic (FigureQA) visual question-answering benchmarks. To the best of our knowledge, this is the first use of self-supervised learning for prompting.
Exponential Moving Average (EMA) is a widely used weight averaging (WA) regularization to learn flat optima for better generalizations without extra cost in deep neural network (DNN) optimization. Despite achieving better flatness, existing WA methods might fall into worse final performances or require extra test-time computations. This work unveils the full potential of EMA with a single line of modification, i.e., switching the EMA parameters to the original model after each epoch, dubbed as Switch EMA (SEMA). From both theoretical and empirical aspects, we demonstrate that SEMA can help DNNs to reach generalization optima that better trade-off between flatness and sharpness. To verify the effectiveness of SEMA, we conduct comparison experiments with discriminative, generative, and regression tasks on vision and language datasets, including image classification, self-supervised learning, object detection and segmentation, image generation, video prediction, attribute regression, and language modeling. Comprehensive results with popular optimizers and networks show that SEMA is a free lunch for DNN training by improving performances and boosting convergence speeds.
Graph Neural Networks (GNNs) have demonstrated their effectiveness in various graph learning tasks, yet their reliance on neighborhood aggregation during inference poses challenges for deployment in latency-sensitive applications, such as real-time financial fraud detection. To address this limitation, recent studies have proposed distilling knowledge from teacher GNNs into student Multi-Layer Perceptrons (MLPs) trained on node content, aiming to accelerate inference. However, these approaches often inadequately explore structural information when inferring unseen nodes. To this end, we introduce SimMLP, a Self-supervised framework for learning MLPs on graphs, designed to fully integrate rich structural information into MLPs. Notably, SimMLP is the first MLP-learning method that can achieve equivalence to GNNs in the optimal case. The key idea is to employ self-supervised learning to align the representations encoded by graph context-aware GNNs and neighborhood dependency-free MLPs, thereby fully integrating the structural information into MLPs. We provide a comprehensive theoretical analysis, demonstrating the equivalence between SimMLP and GNNs based on mutual information and inductive bias, highlighting SimMLP's advanced structural learning capabilities. Additionally, we conduct extensive experiments on 20 benchmark datasets, covering node classification, link prediction, and graph classification, to showcase SimMLP's superiority over state-of-the-art baselines, particularly in scenarios involving unseen nodes (e.g., inductive and cold-start node classification) where structural insights are crucial. Our codes are available at: https://github.com/Zehong-Wang/SimMLP.
In recent years, driven by the need for safer and more autonomous transport systems, the automotive industry has shifted toward integrating a growing number of Advanced Driver Assistance Systems (ADAS). Among the array of sensors employed for object recognition tasks, radar sensors have emerged as a formidable contender due to their abilities in adverse weather conditions or low-light scenarios and their robustness in maintaining consistent performance across diverse environments. However, the small size of radar datasets and the complexity of the labelling of those data limit the performance of radar object detectors. Driven by the promising results of self-supervised learning in computer vision, this paper presents RiCL, an instance contrastive learning framework to pre-train radar object detectors. We propose to exploit the detection from the radar and the temporal information to pre-train the radar object detection model in a self-supervised way using contrastive learning. We aim to pre-train an object detector's backbone, head and neck to learn with fewer data. Experiments on the CARRADA and the RADDet datasets show the effectiveness of our approach in learning generic representations of objects in range-Doppler maps. Notably, our pre-training strategy allows us to use only 20% of the labelled data to reach a similar [email omitted] than a supervised approach using the whole training set.
The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.
Generative self-supervised learning on graphs, particularly graph masked autoencoders, has emerged as a popular learning paradigm and demonstrated its efficacy in handling non-Euclidean data. However, several remaining issues limit the capability of existing methods: 1) the disregard of uneven node significance in masking, 2) the underutilization of holistic graph information, 3) the ignorance of semantic knowledge in the representation space due to the exclusive use of reconstruction loss in the output space, and 4) the unstable reconstructions caused by the large volume of masked contents. In light of this, we propose UGMAE, a unified framework for graph masked autoencoders to address these issues from the perspectives of adaptivity, integrity, complementarity, and consistency. Specifically, we first develop an adaptive feature mask generator to account for the unique significance of nodes and sample informative masks (adaptivity). We then design a ranking-based structure reconstruction objective joint with feature reconstruction to capture holistic graph information and emphasize the topological proximity between neighbors (integrity). After that, we present a bootstrapping-based similarity module to encode the high-level semantic knowledge in the representation space, complementary to the low-level reconstruction in the output space (complementarity). Finally, we build a consistency assurance module to provide reconstruction objectives with extra stabilized consistency targets (consistency). Extensive experiments demonstrate that UGMAE outperforms both contrastive and generative state-of-the-art baselines on several tasks across multiple datasets.
To understand why self-supervised learning (SSL) models have empirically achieved strong performances on several speech-processing downstream tasks, numerous studies have focused on analyzing the encoded information of the SSL layer representations in adult speech. Limited work has investigated how pre-training and fine-tuning affect SSL models encoding children's speech and vocalizations. In this study, we aim to bridge this gap by probing SSL models on two relevant downstream tasks: (1) phoneme recognition (PR) on the speech of adults, older children (8-10 years old), and younger children (1-4 years old), and (2) vocalization classification (VC) distinguishing cry, fuss, and babble for infants under 14 months old. For younger children's PR, the superiority of fine-tuned SSL models is largely due to their ability to learn features that represent older children's speech and then adapt those features to the speech of younger children. For infant VC, SSL models pre-trained on large-scale home recordings learn to leverage phonetic representations at middle layers, and thereby enhance the performance of this task.
Self-supervised learning on graphs can be bifurcated into contrastive and generative methods. Contrastive methods, also known as graph contrastive learning (GCL), have dominated graph self-supervised learning in the past few years, but the recent advent of graph masked autoencoder (GraphMAE) rekindles the momentum behind generative methods. Despite the empirical success of GraphMAE, there is still a dearth of theoretical understanding regarding its efficacy. Moreover, while both generative and contrastive methods have been shown to be effective, their connections and differences have yet to be thoroughly investigated. Therefore, we theoretically build a bridge between GraphMAE and GCL, and prove that the node-level reconstruction objective in GraphMAE implicitly performs context-level GCL. Based on our theoretical analysis, we further identify the limitations of the GraphMAE from the perspectives of alignment and uniformity, which have been considered as two key properties of high-quality representations in GCL. We point out that GraphMAE's alignment performance is restricted by the masking strategy, and the uniformity is not strictly guaranteed. To remedy the aforementioned limitations, we propose an Alignment-Uniformity enhanced Graph Masked AutoEncoder, named AUG-MAE. Specifically, we propose an easy-to-hard adversarial masking strategy to provide hard-to-align samples, which improves the alignment performance. Meanwhile, we introduce an explicit uniformity regularizer to ensure the uniformity of the learned representations. Experimental results on benchmark datasets demonstrate the superiority of our model over existing state-of-the-art methods.
Self-supervised learning (SSL) for automated speech recognition in terms of its emotional content, can be heavily degraded by the presence noise, affecting the efficiency of modeling the intricate temporal and spectral informative structures of speech. Recently, SSL on large speech datasets, as well as new audio-specific SSL proxy tasks, such as, temporal and frequency masking, have emerged, yielding superior performance compared to classic approaches drawn from the image augmentation domain. Our proposed contribution builds upon this successful paradigm by introducing CochCeps-Augment, a novel bio-inspired masking augmentation task for self-supervised contrastive learning of speech representations. Specifically, we utilize the newly introduced bio-inspired cochlear cepstrogram (CCGRAM) to derive noise robust representations of input speech, that are then further refined through a self-supervised learning scheme. The latter employs SimCLR to generate contrastive views of a CCGRAM through masking of its angle and quefrency dimensions. Our experimental approach and validations on the emotion recognition K-EmoCon benchmark dataset, for the first time via a speaker-independent approach, features unsupervised pre-training, linear probing and fine-tuning. Our results potentiate CochCeps-Augment to serve as a standard tool in speech emotion recognition analysis, showing the added value of incorporating bio-inspired masking as an informative augmentation task for self-supervision. Our code for implementing CochCeps-Augment will be made available at: https://github.com/GiannisZgs/CochCepsAugment.
We study the data-generating mechanism for reconstructive SSL to shed light on its effectiveness. With an infinite amount of labeled samples, we provide a sufficient and necessary condition for perfect linear approximation. The condition reveals a full-rank component that preserves the label classes of Y, along with a redundant component. Motivated by the condition, we propose to approximate the redundant component by a low-rank factorization and measure the approximation quality by introducing a new quantity $\epsilon_s$, parameterized by the rank of factorization s. We incorporate $\epsilon_s$ into the excess risk analysis under both linear regression and ridge regression settings, where the latter regularization approach is to handle scenarios when the dimension of the learned features is much larger than the number of labeled samples n for downstream tasks. We design three stylized experiments to compare SSL with supervised learning under different settings to support our theoretical findings.
Medical image segmentation has been significantly advanced by deep learning (DL) techniques, though the data scarcity inherent in medical applications poses a great challenge to DL-based segmentation methods. Self-supervised learning offers a solution by creating auxiliary learning tasks from the available dataset and then leveraging the knowledge acquired from solving auxiliary tasks to help better solve the target segmentation task. Different auxiliary tasks may have different properties and thus can help the target task to different extents. It is desired to leverage their complementary advantages to enhance the overall assistance to the target task. To achieve this, existing methods often adopt a joint training paradigm, which co-solves segmentation and auxiliary tasks by integrating their losses or intermediate gradients. However, direct coupling of losses or intermediate gradients risks undesirable interference because the knowledge acquired from solving each auxiliary task at every training step may not always benefit the target task. To address this issue, we propose a two-stage training approach. In the first stage, the target segmentation task will be independently co-solved with each auxiliary task in both joint training and pre-training modes, with the better model selected via validation performance. In the second stage, the models obtained with respect to each auxiliary task are converted into a single model using an ensemble knowledge distillation method. Our approach allows for making best use of each auxiliary task to create multiple elite segmentation models and then combine them into an even more powerful model. We employed five auxiliary tasks of different proprieties in our approach and applied it to train the U-Net model on an X-ray pneumothorax segmentation dataset. Experimental results demonstrate the superiority of our approach over several existing methods.
Self-supervised Learning (SSL) has emerged as a powerful technique in pre-training deep learning models without relying on expensive annotated labels, instead leveraging embedded signals in unlabeled data. While SSL has shown remarkable success in computer vision tasks through intuitive data augmentation, its application to graph-structured data poses challenges due to the semantic-altering and counter-intuitive nature of graph augmentations. Addressing this limitation, this paper introduces a novel non-contrastive SSL approach to Explicitly Generate a compositional Relation Graph (ExGRG) instead of relying solely on the conventional augmentation-based implicit relation graph. ExGRG offers a framework for incorporating prior domain knowledge and online extracted information into the SSL invariance objective, drawing inspiration from the Laplacian Eigenmap and Expectation-Maximization (EM). Employing an EM perspective on SSL, our E-step involves relation graph generation to identify candidates to guide the SSL invariance objective, and M-step updates the model parameters by integrating the derived relational information. Extensive experimentation on diverse node classification datasets demonstrates the superiority of our method over state-of-the-art techniques, affirming ExGRG as an effective adoption of SSL for graph representation learning.
Given the significance of speech emotion recognition, numerous methods have been developed in recent years to create effective and efficient systems in this domain. One of these methods involves the use of pretrained transformers, fine-tuned to address this specific problem, resulting in high accuracy. Despite extensive discussions and global-scale efforts to enhance these systems, the application of this innovative and effective approach has received less attention in the context of Persian speech emotion recognition. In this article, we review the field of speech emotion recognition and its background, with an emphasis on the importance of employing transformers in this context. We present two models, one based on spectrograms and the other on the audio itself, fine-tuned using the shEMO dataset. These models significantly enhance the accuracy of previous systems, increasing it from approximately 65% to 80% on the mentioned dataset. Subsequently, to investigate the effect of multilinguality on the fine-tuning process, these same models are fine-tuned twice. First, they are fine-tuned using the English IEMOCAP dataset, and then they are fine-tuned with the Persian shEMO dataset. This results in an improved accuracy of 82% for the Persian emotion recognition system. Keywords: Persian Speech Emotion Recognition, shEMO, Self-Supervised Learning
When a mobile robot is asked to navigate intelligently in an environment, it needs to estimate its own and the environment's state. One of the popular methods for robot state and position estimation is particle filtering (PF). Visual Teach and Repeat (VT&R) is a type of navigation that uses a camera to navigate the robot along the previously traversed path. Particle filters are usually used in VT&R to fuse data from odometry and camera to estimate the distance traveled along the path. However, in VT&R, there are other valuable states that the robot can benefit from, especially when moving through changing environments. We propose a multidimensional particle filter to estimate the robot state in VT&R navigation. Apart from the traveled distance, our particle filter estimates lateral and heading deviation from the taught path as well as the current appearance of the environment. This appearance is estimated using maps created across various environmental conditions recorded during the previous traversals. The joint state estimation is based on contrastive neural network architecture, allowing self-supervised learning. This architecture can process multiple images in parallel, alleviating the potential overhead caused by computing the particle filter over the maps simultaneously. We conducted experiments to show that the joint robot/environment state estimation improves navigation accuracy and robustness in a continual mapping setup. Unlike the other frameworks, which treat the robot position and environment appearance separately, our PF represents them as one multidimensional state, resulting in a more general uncertainty model for VT&R.
Due to the complex nature of human emotions and the diversity of emotion representation methods in humans, emotion recognition is a challenging field. In this research, three input modalities, namely text, audio (speech), and video, are employed to generate multimodal feature vectors. For generating features for each of these modalities, pre-trained Transformer models with fine-tuning are utilized. In each modality, a Transformer model is used with transfer learning to extract feature and emotional structure. These features are then fused together, and emotion recognition is performed using a classifier. To select an appropriate fusion method and classifier, various feature-level and decision-level fusion techniques have been experimented with, and ultimately, the best model, which combines feature-level fusion by concatenating feature vectors and classification using a Support Vector Machine on the IEMOCAP multimodal dataset, achieves an accuracy of 75.42%. Keywords: Multimodal Emotion Recognition, IEMOCAP, Self-Supervised Learning, Transfer Learning, Transformer.