Offline policy learning aims to discover decision-making policies from previously-collected datasets without additional online interactions with the environment. As the training dataset is fixed, its quality becomes a crucial determining factor in the performance of the learned policy. This paper studies a dataset characteristic that we refer to as multi-behavior, indicating that the dataset is collected using multiple policies that exhibit distinct behaviors. In contrast, a uni-behavior dataset would be collected solely using one policy. We observed that policies learned from a uni-behavior dataset typically outperform those learned from multi-behavior datasets, despite the uni-behavior dataset having fewer examples and less diversity. Therefore, we propose a behavior-aware deep clustering approach that partitions multi-behavior datasets into several uni-behavior subsets, thereby benefiting downstream policy learning. Our approach is flexible and effective; it can adaptively estimate the number of clusters while demonstrating high clustering accuracy, achieving an average Adjusted Rand Index of 0.987 across various continuous control task datasets. Finally, we present improved policy learning examples using dataset clustering and discuss several potential scenarios where our approach might benefit the offline policy learning community.

Video Anomaly Detection (VAD) is an open-set recognition task, which is usually formulated as a one-class classification (OCC) problem, where training data is comprised of videos with normal instances while test data contains both normal and anomalous instances. Recent works have investigated the creation of pseudo-anomalies (PAs) using only the normal data and making strong assumptions about real-world anomalies with regards to abnormality of objects and speed of motion to inject prior information about anomalies in an autoencoder (AE) based reconstruction model during training. This work proposes a novel method for generating generic spatio-temporal PAs by inpainting a masked out region of an image using a pre-trained Latent Diffusion Model and further perturbing the optical flow using mixup to emulate spatio-temporal distortions in the data. In addition, we present a simple unified framework to detect real-world anomalies under the OCC setting by learning three types of anomaly indicators, namely reconstruction quality, temporal irregularity and semantic inconsistency. Extensive experiments on four VAD benchmark datasets namely Ped2, Avenue, ShanghaiTech and UBnormal demonstrate that our method performs on par with other existing state-of-the-art PAs generation and reconstruction based methods under the OCC setting. Our analysis also examines the transferability and generalisation of PAs across these datasets, offering valuable insights by identifying real-world anomalies through PAs.

Hindsight Experience Replay (HER) is a technique used in reinforcement learning (RL) that has proven to be very efficient for training off-policy RL-based agents to solve goal-based robotic manipulation tasks using sparse rewards. Even though HER improves the sample efficiency of RL-based agents by learning from mistakes made in past experiences, it does not provide any guidance while exploring the environment. This leads to very large training times due to the volume of experience required to train an agent using this replay strategy. In this paper, we propose a method that uses primitive behaviours that have been previously learned to solve simple tasks in order to guide the agent toward more rewarding actions during exploration while learning other more complex tasks. This guidance, however, is not executed by a manually designed curriculum, but rather using a critic network to decide at each timestep whether or not to use the actions proposed by the previously-learned primitive policies. We evaluate our method by comparing its performance against HER and other more efficient variations of this algorithm in several block manipulation tasks. We demonstrate the agents can learn a successful policy faster when using our proposed method, both in terms of sample efficiency and computation time. Code is available at https://github.com/franroldans/qmp-her.

Learning control policies offline from pre-recorded datasets is a promising avenue for solving challenging real-world problems. However, available datasets are typically of mixed quality, with a limited number of the trajectories that we would consider as positive examples; i.e., high-quality demonstrations. Therefore, we propose a novel iterative learning algorithm for identifying expert trajectories in unlabeled mixed-quality robotics datasets given a minimal set of positive examples, surpassing existing algorithms in terms of accuracy. We show that applying behavioral cloning to the resulting filtered dataset outperforms several competitive offline reinforcement learning and imitation learning baselines. We perform experiments on a range of simulated locomotion tasks and on two challenging manipulation tasks on a real robotic system; in these experiments, our method showcases state-of-the-art performance. Our website: \url{https://sites.google.com/view/offline-policy-learning-pubc}.

This paper presents our solution for the Real Robot Challenge (RRC) III, a competition featured in the NeurIPS 2022 Competition Track, aimed at addressing dexterous robotic manipulation tasks through learning from pre-collected offline data. Participants were provided with two types of datasets for each task: expert and mixed datasets with varying skill levels. While the simplest offline policy learning algorithm, Behavioral Cloning (BC), performed remarkably well when trained on expert datasets, it outperformed even the most advanced offline reinforcement learning (RL) algorithms. However, BC's performance deteriorated when applied to mixed datasets, and the performance of offline RL algorithms was also unsatisfactory. Upon examining the mixed datasets, we observed that they contained a significant amount of expert data, although this data was unlabeled. To address this issue, we proposed a semi-supervised learning-based classifier to identify the underlying expert behavior within mixed datasets, effectively isolating the expert data. To further enhance BC's performance, we leveraged the geometric symmetry of the RRC arena to augment the training dataset through mathematical transformations. In the end, our submission surpassed that of all other participants, even those who employed complex offline RL algorithms and intricate data processing and feature engineering techniques.

Self-supervised representation learning methods [4, 7, 11, 12, 14] are the standard approach for training large scale deep neural networks (DNNs). One of the main reasons for their popularity is their capability to leverage the inherent structure of data from a vast unlabeled corpus during pre-training, which makes them highly suitable for transfer learning [28]. However, this comes at the cost of substantially larger model size, computationally expensive training strategies (larger training times, large batch-sizes, etc.) [13, 28] and subsequently more expensive inference times. Though such strategies are effective for achieving state-of-the-art results in computer vision, they may not be practical in resource-constrained industrial settings that require lightweight models to be deployed on edge devices. To lessen the computational burden, it is common to extract (or learn) a lightweight network from an off-the-shelf pre-trained model. This has been successfully achieved through techniques such as knowledge distillation (KD) [35], pruning [24], dynamic computation (DC) [58], etc. KD methods follow a standard two-step procedure of pre-training and distilling knowledge into a student network using self-supervised (SS) objective [1, 21, 51] or by together incorporating supervised and SS objectives [54], while pruning based approaches heavily rely on multiple steps of pre-train prune finetune to get a lightweight network irrespective of the objective, whereas methods based on dynamic/conditional computation [34, 58] again rely on a pre-trained model to obtain a lightweight network while keeping the network topology intact via a gating mechanism. These approaches are effective but using fine-tuning to obtain a sub-network from large pre-trained models (such as Large Language Models) can be computationally expensive and cumbersome. Also, since downstream tasks are diverse and vary widely, any change in the task requires repeating the entire procedure multiple times, making it inefficient and less transferable.

Contrastive pretrained large Vision-Language Models (VLMs) like CLIP have revolutionized visual representation learning by providing good performance on downstream datasets. VLMs are 0-shot adapted to a downstream dataset by designing prompts that are relevant to the dataset. Such prompt engineering makes use of domain expertise and a validation dataset. Meanwhile, recent developments in generative pretrained models like GPT-4 mean they can be used as advanced internet search tools. They can also be manipulated to provide visual information in any structure. In this work, we show that GPT-4 can be used to generate text that is visually descriptive and how this can be used to adapt CLIP to downstream tasks. We show considerable improvements in 0-shot transfer accuracy on specialized fine-grained datasets like EuroSAT (~7%), DTD (~7%), SUN397 (~4.6%), and CUB (~3.3%) when compared to CLIP's default prompt. We also design a simple few-shot adapter that learns to choose the best possible sentences to construct generalizable classifiers that outperform the recently proposed CoCoOP by ~2% on average and by over 4% on 4 specialized fine-grained datasets. The code, prompts, and auxiliary text dataset is available at https://github.com/mayug/VDT-Adapter.

Deep learning has shown excellent performance in analysing medical images. However, datasets are difficult to obtain due privacy issues, standardization problems, and lack of annotations. We address these problems by producing realistic synthetic images using a combination of 3D technologies and generative adversarial networks. We propose CUT-seg, a joint training where a segmentation model and a generative model are jointly trained to produce realistic images while learning to segment polyps. We take advantage of recent one-sided translation models because they use significantly less memory, allowing us to add a segmentation model in the training loop. CUT-seg performs better, is computationally less expensive, and requires less real images than other memory-intensive image translation approaches that require two stage training. Promising results are achieved on five real polyp segmentation datasets using only one real image and zero real annotations. As a part of this study we release Synth-Colon, an entirely synthetic dataset that includes 20000 realistic colon images and additional details about depth and 3D geometry: https://enric1994.github.io/synth-colon

COVID-19 has created a global health crisis with millions of cases and deaths reported worldwide. Timely and accurate COVID severity prediction is essential for effective clinical management and treatment. Deep learning has shown tremendous potential in the medical domain. Automating the severity prediction of COVID-19 based on deep learning can lead to improved clinical workflow, resulting in faster diagnosis and better prognosis for severe COVID-19 cases. While a large number of studies have utilized deep learning methods for COVID-19 prediction based on chest X-ray data, CT scans have been found to be more effective in detecting COVID-19 positivity and severity [Aswathy et al., 2021]. Previous studies have primarily focused on COVID-19 positivity prediction or improving feature extraction methods. However, limited work has been done on COVID-19 severity prediction using CT scans, which are valuable for assessing lung condition, predicting COVID-19 severity, and detecting complications.

This paper describes deep learning models based on convolutional neural networks applied to the problem of predicting EM wave propagation over rural terrain. A surface integral equation formulation, solved with the method of moments and accelerated using the Fast Far Field approximation, is used to generate synthetic training data which comprises path loss computed over randomly generated 1D terrain profiles. These are used to train two networks, one based on fractal profiles and one based on profiles generated using a Gaussian process. The models show excellent agreement when applied to test profiles generated using the same statistical process used to create the training data and very good accuracy when applied to real life problems.