We propose to learn to distinguish reversible from irreversible actions for better informed decision-making in Reinforcement Learning (RL). From theoretical considerations, we show that approximate reversibility can be learned through a simple surrogate task: ranking randomly sampled trajectory events in chronological order. Intuitively, pairs of events that are always observed in the same order are likely to be separated by an irreversible sequence of actions. Conveniently, learning the temporal order of events can be done in a fully self-supervised way, which we use to estimate the reversibility of actions from experience, without any priors.We propose two different strategies that incorporate reversibility in RL agents, one strategy for exploration (RAE) and one strategy for control (RAC). We demonstrate the potential of reversibility-aware agents in several environments, including the challenging Sokoban game. In synthetic tasks, we show that we can learn control policies that never fail and reduce to zero the side-effects of interactions, even without access to the reward function.

Speech Emotion Recognition (SER) has seen significant progress with deep learning, yet remains challenging for Low-Resource Languages (LRLs) due to the scarcity of annotated data. In this work, we explore unsupervised learning to improve SER in low-resource settings. Specifically, we investigate contrastive learning (CL) and Bootstrap Y our Own Latent (BYOL) as self-supervised approaches to enhance cross-lingual generalization. Our methods achieve notable F1 score improvements of 10.6% in Urdu, 15.2% in German, and 13.9% in Bangla, demonstrating their effectiveness in LRLs. Additionally, we analyze model behavior to provide insights on key factors influencing performance across languages, and also highlighting challenges in low-resource SER. This work provides a foundation for developing more inclusive, explainable, and robust emotion recognition systems for underrepresented languages.

Mobile robots rely on maps to navigate through an environment. In the absence of any map, the robots must build the map online from partial observations as they move in the environment. Traditional methods build a map using only direct observations. In contrast, humans identify patterns in the observed environment and make informed guesses about what to expect ahead. Modeling these patterns explicitly is difficult due to the complexity of the environments. However, these complex models can be approximated well using learning-based methods in conjunction with large training data. By extracting patterns, robots can use direct observations and predictions of what lies ahead to better navigate an unknown environment. In this dissertation, we present several learning-based methods to equip mobile robots with prediction capabilities for efficient and safer operation. In the first part of the dissertation, we learn to predict using geometrical and structural patterns in the environment. Partially observed maps provide invaluable cues for accurately predicting the unobserved areas. We first demonstrate the capability of general learning-based approaches to model these patterns for a variety of overhead map modalities. Then we employ task-specific learning for faster navigation in indoor environments by predicting 2D occupancy in the nearby regions. This idea is further extended to 3D point cloud representation for object reconstruction. Predicting the shape of the full object from only partial views, our approach paves the way for efficient next-best-view planning. In the second part of the dissertation, we learn to predict using spatiotemporal patterns in the environment. We focus on dynamic tasks such as target tracking and coverage where we seek decentralized coordination between robots. We first show how graph neural networks can be used for more scalable and faster inference.

We propose to learn to distinguish reversible from irreversible actions for better informed decision-making in Reinforcement Learning (RL). From theoretical considerations, we show that approximate reversibility can be learned through a simple surrogate task: ranking randomly sampled trajectory events in chronological order. Intuitively, pairs of events that are always observed in the same order are likely to be separated by an irreversible sequence of actions. Conveniently, learning the temporal order of events can be done in a fully self-supervised way, which we use to estimate the reversibility of actions from experience, without any priors.We propose two different strategies that incorporate reversibility in RL agents, one strategy for exploration (RAE) and one strategy for control (RAC). We demonstrate the potential of reversibility-aware agents in several environments, including the challenging Sokoban game.

Computer vision in agriculture is game-changing with its ability to transform farming into a data-driven, precise, and sustainable industry. Deep learning has empowered agriculture vision to analyze vast, complex visual data, but heavily rely on the availability of large annotated datasets. This remains a bottleneck as manual labeling is error-prone, time-consuming, and expensive. The lack of efficient labeling approaches inspired us to consider self-supervised learning as a paradigm shift, learning meaningful feature representations from raw agricultural image data. In this work, we explore how self-supervised representation learning unlocks the potential applicability to diverse agriculture vision tasks by eliminating the need for large-scale annotated datasets. We propose a lightweight framework utilizing SimCLR, a contrastive learning approach, to pre-train a ResNet-50 backbone on a large, unannotated dataset of real-world agriculture field images. Our experimental analysis and results indicate that the model learns robust features applicable to a broad range of downstream agriculture tasks discussed in the paper. Additionally, the reduced reliance on annotated data makes our approach more cost-effective and accessible, paving the way for broader adoption of computer vision in agriculture.

Recent advancements in self-supervised learning have unlocked the potential to harness unlabeled data for auxiliary tasks, facilitating the learning of beneficial priors. This has been particularly advantageous in fields like medical image analysis, where labeled data are scarce. Although effective for classification tasks, this methodology has shown limitations in more complex applications, such as medical image segmentation. In this paper, we introduce Medical imaging Enhanced with Dynamic Self-Adaptive Semantic Segmentation (MedSASS), a dedicated self-supervised framework tailored for medical image segmentation. We evaluate MedSASS against existing state-of-the-art methods across four diverse medical datasets, showcasing its superiority. MedSASS outperforms existing CNN-based self-supervised methods by 3.83% and matches the performance of ViT-based methods. Furthermore, when MedSASS is trained end-to-end, covering both encoder and decoder, it demonstrates significant improvements of 14.4% for CNNs and 6% for ViT-based architectures compared to existing state-of-the-art self-supervised strategies.

This paper addresses the problem of global tempo estimation in musical audio. Given that annotating tempo is time-consuming and requires certain musical expertise, few publicly available data sources exist to train machine learning models for this task. Towards alleviating this issue, we propose a fully self-supervised approach that does not rely on any human labeled data. Our method builds on the fact that generic (music) audio embeddings already encode a variety of properties, including information about tempo, making them easily adaptable for downstream tasks. While recent work in self-supervised tempo estimation aimed to learn a tempo specific representation that was subsequently used to train a supervised classifier, we reformulate the task into the binary classification problem of predicting whether a target track has the same or a different tempo compared to a reference. While the former still requires labeled training data for the final classification model, our approach uses arbitrary unlabeled music data in combination with time-stretching for model training as well as a small set of synthetically created reference samples for predicting the final tempo. Evaluation of our approach in comparison with the state-of-the-art reveals highly competitive performance when the constraint of finding the precise tempo octave is relaxed.

The process of annotating relevant data in the field of digital microscopy can be both time-consuming and especially expensive due to the required technical skills and human-expert knowledge. Consequently, large amounts of microscopic image data sets remain unlabeled, preventing their effective exploitation using deep-learning algorithms. In recent years it has been shown that a lot of relevant information can be drawn from unlabeled data. Self-supervised learning (SSL) is a promising solution based on learning intrinsic features under a pretext task that is similar to the main task without requiring labels. The trained result is transferred to the main task - image segmentation in our case. A ResNet50 U-Net was first trained to restore images of liver progenitor organoids from augmented images using the Structural Similarity Index Metric (SSIM), alone, and using SSIM combined with L1 loss. Both the encoder and decoder were trained in tandem. The weights were transferred to another U-Net model designed for segmentation with frozen encoder weights, using Binary Cross Entropy, Dice, and Intersection over Union (IoU) losses. For comparison, we used the same U-Net architecture to train two supervised models, one utilizing the ResNet50 encoder as well as a simple CNN. Results showed that self-supervised learning models using a 25\% pixel drop or image blurring augmentation performed better than the other augmentation techniques using the IoU loss. When trained on only 114 images for the main task, the self-supervised learning approach outperforms the supervised method achieving an F1-score of 0.85, with higher stability, in contrast to an F1=0.78 scored by the supervised method. Furthermore, when trained with larger data sets (1,000 images), self-supervised learning is still able to perform better, achieving an F1-score of 0.92, contrasting to a score of 0.85 for the supervised method.

Document layout analysis is a known problem to the documents research community and has been vastly explored yielding a multitude of solutions ranging from text mining, and recognition to graph-based representation, visual feature extraction, etc. However, most of the existing works have ignored the crucial fact regarding the scarcity of labeled data. With growing internet connectivity to personal life, an enormous amount of documents had been available in the public domain and thus making data annotation a tedious task. We address this challenge using self-supervision and unlike, the few existing self-supervised document segmentation approaches which use text mining and textual labels, we use a complete vision-based approach in pre-training without any ground-truth label or its derivative. Instead, we generate pseudo-layouts from the document images to pre-train an image encoder to learn the document object representation and localization in a self-supervised framework before fine-tuning it with an object detection model. We show that our pipeline sets a new benchmark in this context and performs at par with the existing methods and the supervised counterparts, if not outperforms. The code is made publicly available at: https://github.com/MaitySubhajit/SelfDocSeg

Maintaining proper form while exercising is important for preventing injuries and maximizing muscle mass gains. Detecting errors in workout form naturally requires estimating human's body pose. However, off-the-shelf pose estimators struggle to perform well on the videos recorded in gym scenarios due to factors such as camera angles, occlusion from gym equipment, illumination, and clothing. To aggravate the problem, the errors to be detected in the workouts are very subtle. To that end, we propose to learn exercise-oriented image and video representations from unlabeled samples such that a small dataset annotated by experts suffices for supervised error detection. In particular, our domain knowledge-informed self-supervised approaches (pose contrastive learning and motion disentangling) exploit the harmonic motion of the exercise actions, and capitalize on the large variances in camera angles, clothes, and illumination to learn powerful representations. To facilitate our self-supervised pretraining, and supervised finetuning, we curated a new exercise dataset, \emph{Fitness-AQA} (\url{https://github.com/ParitoshParmar/Fitness-AQA}), comprising of three exercises: BackSquat, BarbellRow, and OverheadPress. It has been annotated by expert trainers for multiple crucial and typically occurring exercise errors. Experimental results show that our self-supervised representations outperform off-the-shelf 2D- and 3D-pose estimators and several other baselines. We also show that our approaches can be applied to other domains/tasks such as pose estimation and dive quality assessment.