Search

Self-supervised representation learning (SSRL) methods aim to provide powerful, deep feature learning without the requirement of large annotated data sets, thus alleviating the annotation bottleneck—one of the main barriers to the practical deployment of deep learning today. These techniques have advanced rapidly in recent years, with their efficacy approaching and sometimes surpassing fully supervised pretraining alternatives across a variety of data modalities, including image, video, sound, text, and graphs. This article introduces this vibrant area, including key concepts, the four main families of approaches and associated state-of-the-art techniques, and how self-supervised methods are applied to diverse modalities of data. We further discuss practical considerations including workflows, representation transferability, and computational cost. Finally, we survey major open challenges in the field, that provide fertile ground for future work.

We present a "learning to learn" approach for automatically constructing white-box classification loss functions that are robust to label noise in the training data. We parameterize a flexible family of loss functions using Taylor polynomials, and apply evolutionary strategies to search for noise-robust losses in this space. To learn re-usable loss functions that can apply to new tasks, our fitness function scores their performance in aggregate across a range of training dataset and architecture combinations. The resulting white-box loss provides a simple and fast "plug-and-play" module that enables effective noise-robust learning in diverse downstream tasks, without requiring a special training procedure or network architecture. The efficacy of our method is demonstrated on a variety of datasets with both synthetic and real label noise, where we compare favourably to previous work.

Optimisers are an essential component for training machine learning models, and their design influences learning speed and generalisation. Several studies have attempted to learn more effective gradient-descent optimisers via solving a bi-level optimisation problem where generalisation error is minimised with respect to optimiser parameters. However, most existing optimiser learning methods are intuitively motivated, without clear theoretical support. We take a different perspective starting from mirror descent rather than gradient descent, and meta-learning the corresponding Bregman divergence. Within this paradigm, we formalise a novel meta-learning objective of minimising the regret bound of learning. The resulting framework, termed Meta Mirror Descent (MetaMD), learns to accelerate optimisation speed. Unlike many meta-learned optimisers, it also supports convergence and generalisation guarantees and uniquely does so without requiring validation data. We evaluate our framework on a variety of tasks and architectures in terms of convergence rate and generalisation error and demonstrate strong performance.

Recent sharpness-aware minimisation (SAM) is known to find flat minima which is beneficial for better generalisation with improved robustness. SAM essentially modifies the loss function by reporting the maximum loss value within the small neighborhood around the current iterate. However, it uses the Euclidean ball to define the neighborhood, which can be inaccurate since loss functions for neural networks are typically defined over probability distributions (e.g., class predictive probabilities), rendering the parameter space non Euclidean. In this paper we consider the information geometry of the model parameter space when defining the neighborhood, namely replacing SAM's Euclidean balls with ellipsoids induced by the Fisher information. Our approach, dubbed Fisher SAM, defines more accurate neighborhood structures that conform to the intrinsic metric of the underlying statistical manifold. For instance, SAM may probe the worst-case loss value at either a too nearby or inappropriately distant point due to the ignorance of the parameter space geometry, which is avoided by our Fisher SAM. Another recent Adaptive SAM approach stretches/shrinks the Euclidean ball in accordance with the scale of the parameter magnitudes. This might be dangerous, potentially destroying the neighborhood structure. We demonstrate improved performance of the proposed Fisher SAM on several benchmark datasets/tasks.

Federated learning (FL) enables distributed participants to collectively learn a strong global model without sacrificing their individual data privacy. Mainstream FL approaches require each participant to share a common network architecture and further assume that data are are sampled IID across participants. However, in real-world deployments participants may require heterogeneous network architectures; and the data distribution is almost certainly non-uniform across participants. To address these issues we introduce FedH2L, which is agnostic to both the model architecture and robust to different data distributions across participants. In contrast to approaches sharing parameters or gradients, FedH2L relies on mutual distillation, exchanging only posteriors on a shared seed set between participants in a decentralized manner. This makes it extremely bandwidth efficient, model agnostic, and crucially produces models capable of performing well on the whole data distribution when learning from heterogeneous silos.

In the absence of large labelled datasets, self-supervised learning techniques can boost performance by learning useful representations from unlabelled data, which is often more readily available. However, there is often a domain shift between the unlabelled collection and the downstream target problem data. We show that by learning Bayesian instance weights for the unlabelled data, we can improve the downstream classification accuracy by prioritising the most useful instances. Additionally, we show that the training time can be reduced by discarding unnecessary datapoints. Our method, BetaDataWeighter is evaluated using the popular self-supervised rotation prediction task on STL-10 and Visual Decathlon. We compare to related instance weighting schemes, both hand-designed heuristics and meta-learning, as well as conventional self-supervised learning. BetaDataWeighter achieves both the highest average accuracy and rank across datasets, and on STL-10 it prunes up to 78% of unlabelled images without significant loss in accuracy, corresponding to over 50% reduction in training time.

Off-Policy Actor-Critic (Off-PAC) methods have proven successful in a variety of continuous control tasks. Normally, the critic's action-value function is updated using temporal-difference, and the critic in turn provides a loss for the actor that trains it to take actions with higher expected return. In this paper, we introduce a novel and flexible meta-critic that observes the learning process and meta-learns an additional loss for the actor that accelerates and improves actor-critic learning. Compared to the vanilla critic, the meta-critic network is explicitly trained to accelerate the learning process; and compared to existing meta-learning algorithms, meta-critic is rapidly learned online for a single task, rather than slowly over a family of tasks. Crucially, our meta-critic framework is designed for off-policy based learners, which currently provide state-of-the-art reinforcement learning sample efficiency. We demonstrate that online meta-critic learning leads to improvements in avariety of continuous control environments when combined with contemporary Off-PAC methods DDPG, TD3 and the state-of-the-art SAC., Comment: NeurIPS 2020

We investigate approaches to regularisation during fine-tuning of deep neural networks. First we provide a neural network generalisation bound based on Rademacher complexity that uses the distance the weights have moved from their initial values. This bound has no direct dependence on the number of weights and compares favourably to other bounds when applied to convolutional networks. Our bound is highly relevant for fine-tuning, because providing a network with a good initialisation based on transfer learning means that learning can modify the weights less, and hence achieve tighter generalisation. Inspired by this, we develop a simple yet effective fine-tuning algorithm that constrains the hypothesis class to a small sphere centred on the initial pre-trained weights, thus obtaining provably better generalisation performance than conventional transfer learning. Empirical evaluation shows that our algorithm works well, corroborating our theoretical results. It outperforms both state of the art fine-tuning competitors, and penalty-based alternatives that we show do not directly constrain the radius of the search space.

We address the problem of simultaneously learning a k-means clustering and deep feature representation from unlabelled data, which is of interest due to the potential of deep k-means to outperform traditional two-step feature extraction and shallow-clustering strategies. We achieve this by developing a gradient-estimator for the non-differentiable k-means objective via the Gumbel-Softmax reparameterisation trick. In contrast to previous attempts at deep clustering, our concrete k-means model can be optimised with respect to the canonical k-means objective and is easily trained end-to-end without resorting to alternating optimisation. We demonstrate the efficacy of our method on standard clustering benchmarks.

The well known domain shift issue causes model performance to degrade when deployed to a new target domain with different statistics to training. Domain adaptation techniques alleviate this, but need some instances from the target domain to drive adaptation. Domain generalisation is the recently topical problem of learning a model that generalises to unseen domains out of the box, and various approaches aim to train a domain-invariant feature extractor, typically by adding some manually designed losses. In this work, we propose a learning to learn approach, where the auxiliary loss that helps generalisation is itself learned. Beyond conventional domain generalisation, we consider a more challenging setting of heterogeneous domain generalisation, where the unseen domains do not share label space with the seen ones, and the goal is to train a feature representation that is useful off-the-shelf for novel data and novel categories. Experimental evaluation demonstrates that our method outperforms state-of-the-art solutions in both settings., Presented at ICML 2019

Adversarial examples are of wide concern due to their impact on the reliability of contemporary machine learning systems. Effective adversarial examples are mostly found via white-box attacks. However, in some cases they can be transferred across models, thus enabling them to attack black-box models. In this work we evaluate the transferability of three adversarial attacks - the Fast Gradient Sign Method, the Basic Iterative Method, and the Carlini & Wagner method, across two classes of models - the VGG class(using VGG16, VGG19 and an ensemble of VGG16 and VGG19), and the Inception class(Inception V3, Xception, Inception Resnet V2, and an ensemble of the three). We also outline the problems with the assessment of transferability in the current body of research and attempt to amend them by picking specific "strong" parameters for the attacks, and by using a L-Infinity clipping technique and the SSIM metric for the final evaluation of the attack transferability.

Deep Reinforcement Learning has shown great success in a variety of control tasks. However, it is unclear how close we are to the vision of putting Deep RL into practice to solve real world problems. In particular, common practice in the field is to train policies on largely deterministic simulators and to evaluate algorithms through training performance alone, without a train/test distinction to ensure models generalise and are not overfitted. Moreover, it is not standard practice to check for generalisation under domain shift, although robustness to such system change between training and testing would be necessary for real-world Deep RL control, for example, in robotics. In this paper we study these issues by first characterising the sources of uncertainty that provide generalisation challenges in Deep RL. We then provide a new benchmark and thorough empirical evaluation of generalisation challenges for state of the art Deep RL methods. In particular, we show that, if generalisation is the goal, then common practice of evaluating algorithms based on their training performance leads to the wrong conclusions about algorithm choice. Finally, we evaluate several techniques for improving generalisation and draw conclusions about the most robust techniques to date.

Deep neural networks have been proven powerful at processing perceptual data, such as images and audio. However for tabular data, tree-based models are more popular. A nice property of tree-based models is their natural interpretability. In this work, we present Deep Neural Decision Trees (DNDT) -- tree models realised by neural networks. A DNDT is intrinsically interpretable, as it is a tree. Yet as it is also a neural network (NN), it can be easily implemented in NN toolkits, and trained with gradient descent rather than greedy splitting. We evaluate DNDT on several tabular datasets, verify its efficacy, and investigate similarities and differences between DNDT and vanilla decision trees. Interestingly, DNDT self-prunes at both split and feature-level., Comment: presented at 2018 ICML Workshop on Human Interpretability in Machine Learning (WHI 2018), Stockholm, Sweden

Multi-domain learning aims to benefit from simultaneously learning across several different but related domains. In this chapter, we propose a single framework that unifies multi-domain learning (MDL) and the related but better studied area of multi-task learning (MTL). By exploiting the concept of a \emph{semantic descriptor} we show how our framework encompasses various classic and recent MDL/MTL algorithms as special cases with different semantic descriptor encodings. As a second contribution, we present a higher order generalisation of this framework, capable of simultaneous multi-task-multi-domain learning. This generalisation has two mathematically equivalent views in multi-linear algebra and gated neural networks respectively. Moreover, by exploiting the semantic descriptor, it provides neural networks the capability of zero-shot learning (ZSL), where a classifier is generated for an unseen class without any training data; as well as zero-shot domain adaptation (ZSDA), where a model is generated for an unseen domain without any training data. In practice, this framework provides a powerful yet easy to implement method that can be flexibly applied to MTL, MDL, ZSL and ZSDA., Invited book chapter

We propose a framework for training multiple neural networks simultaneously. The parameters from all models are regularised by the tensor trace norm, so that each neural network is encouraged to reuse others' parameters if possible -- this is the main motivation behind multi-task learning. In contrast to many deep multi-task learning models, we do not predefine a parameter sharing strategy by specifying which layers have tied parameters. Instead, our framework considers sharing for all shareable layers, and the sharing strategy is learned in a data-driven way., Comment: Submission to Workshop track - ICLR 2017

Recently, zero-shot learning (ZSL) has received increasing interest. The key idea underpinning existing ZSL approaches is to exploit knowledge transfer via an intermediate-level semantic representation which is assumed to be shared between the auxiliary and target datasets, and is used to bridge between these domains for knowledge transfer. The semantic representation used in existing approaches varies from visual attributes to semantic word vectors and semantic relatedness. However, the overall pipeline is similar: a projection mapping low-level features to the semantic representation is learned from the auxiliary dataset by either classification or regression models and applied directly to map each instance into the same semantic representation space where a zero-shot classifier is used to recognise the unseen target class instances with a single known 'prototype' of each target class. In this paper we discuss two related lines of work improving the conventional approach: exploiting transductive learning ZSL, and generalising ZSL to the multi-label case., 4 pages, 4 figures, ECCV 2014 Workshop on Parts and Attributes

In this paper, we provide a new neural-network based perspective on multi-task learning (MTL) and multi-domain learning (MDL). By introducing the concept of a semantic descriptor, this framework unifies MDL and MTL as well as encompassing various classic and recent MTL/MDL algorithms by interpreting them as different ways of constructing semantic descriptors. Our interpretation provides an alternative pipeline for zero-shot learning (ZSL), where a model for a novel class can be constructed without training data. Moreover, it leads to a new and practically relevant problem setting of zero-shot domain adaptation (ZSDA), which is the analogous to ZSL but for novel domains: A model for an unseen domain can be generated by its semantic descriptor. Experiments across this range of problems demonstrate that our framework outperforms a variety of alternatives., 9 pages, Accepted to ICLR 2015 Conference Track