Approaches to class-incremental learning aim to successfully learn from continuously arriving classes. One added level of difficulty usually arises when the training data belonging to each class is scarce, which is the case in several open-world machine learning applications. In this paradigm, which is referred to as few-shot class-incremental learning, a typical learner needs to both be able to learn incrementally from the sequentially arriving classes, and preserve the knowledge which already exists about the old (i.e. already existing) classes. We propose a few-shot class-incremental learner which adapts the representations of the new few-shot classes as well as relevant previous knowledge based on a latent variable model. The proposed latent variable model is a form of a variational autoencoder that is designed to address the main challenges of the few-shot class-incremental learning paradigm, namely catastrophic forgetting and potential bias. During the few-shot learning of new classes, the amortization and high fidelity characteristics of the proposed model are leveraged to adapt not only the current class, but also the relevant previously encountered classes, in order to consistently mitigate the impact of catastrophic forgetting, bias and overfitting. We also derive a generalization upper bound on the error of an upcoming class. Experiments on several widely used few-shot class-incremental learning benchmarks, as well as a medical benchmark consisting of real-world medical images, demonstrate that the proposed model leads to improved performance, as measured by average overall and final classification accuracy, and in terms of alleviating catastrophic forgetting.

Continual learning (CL) is a paradigm which addresses the issue of how to learn from sequentially arriving tasks. The goal of this paper is to introduce a CL framework which can both learn from a global multi-task architecture and locally adapt this learning to the task at hand. In addition to the global knowledge, we conjecture that it is also beneficial to further focus on the most relevant pieces of previous knowledge. Using a prototypical network as a proxy, the proposed framework bases its adaptation on the similarity between the current data stream and the previously encountered data. We develop two algorithms, one for the standard task-aware CL and another for the more challenging task-free setting where boundaries between tasks are unknown. We correspondingly derive a generalization upper bound on the error of an upcoming task. Experiments demonstrate that the introduced algorithms lead to improved performance on several CL benchmarks.

Meta-learning models learn to generalise to unseen tasks at test time.We introduce a meta-learning algorithm which balances (global) generalisation with a (local) adaptive mechanism allowing the meta-learner to deal with potentially substantial heterogeneity in the task distribution.The proposed meta-learner flexibly consolidates shared components (responsible for generalisation) with task-specific components.The latter components are adapted, in a data-driven manner, based on estimating the similarity between the meta-test task in hand and the training tasks.Experiments demonstrate improved performance on few-shot learning benchmarks, both general and others involving a more heterogeneous set of tasks.

There is currently a great expansion of the impact of machine learning algorithms on our lives, prompting the need for objectives other than pure performance, including fairness. Fairness here means that the outcome of an automated decisionmaking system should not discriminate between subgroups characterized by sensitive attributes such as gender or race. Given any existing differentiable classifier, we make only slight adjustments to the architecture including adding a new hidden layer, in order to enable the concurrent adversarial optimization for fairness and accuracy. Our framework provides one way to quantify the tradeoff between fairness and accuracy, while also leading to strong empirical performance.

One of the challenges to reinforcement learning (RL) is scalable transferability among complex tasks. Incorporating a graphical model (GM), along with the rich family of related methods, as a basis for RL frameworks provides potential to address issues such as transferability, generalisation and exploration. Here we propose a flexible GM-based RL framework which leverages efficient inference procedures to enhance generalisation and transfer power. In our proposed transferable and information-based graphical model framework ‘TibGM’, we show the equivalence between our mutual information-based objective in the GM, and an RL consolidated objective consisting of a standard reward maximisation target and a generalisation/transfer objective. In settings where there is a sparse or deceptive reward signal, our TibGM framework is flexible enough to incorporate exploration bonuses depicting intrinsic rewards. We empirically verify improved performance and exploration power.

Interpretability of representations in both deep generative and discriminative models is highly desirable. Current methods jointly optimize an objective combining accuracy and interpretability. However, this may reduce accuracy, and is not applicable to already trained models. We propose two interpretability frameworks. First, we provide an interpretable lens for an existing model. We use a generative model which takes as input the representation in an existing (generative or discriminative) model, weakly supervised by limited side information. Applying a flexible and invertible transformation to the input leads to an interpretable representation with no loss in accuracy. We extend the approach using an active learning strategy to choose the most useful side information to obtain, allowing a human to guide what "interpretable" means. Our second framework relies on joint optimization for a representation which is both maximally informative about the side information and maximally compressive about the non-interpretable data factors. This leads to a novel perspective on the relationship between compression and regularization. We also propose a new interpretability evaluation metric based on our framework. Empirically, we achieve state-of-the-art results on three datasets using the two proposed algorithms.

We present new algorithms for learning Bayesian networks from data with missing values using a data augmentation approach. An exact Bayesian network learning algorithm is obtained by recasting the problem into a standard Bayesian network learning problem without missing data. As expected, the exact algorithm does not scale to large domains. We build on the exact method to create an approximate algorithm using a hill-climbing technique. This algorithm scales to large domains so long as a suitable standard structure learning method for complete data is available. We perform a wide range of experiments to demonstrate the benefits of learning Bayesian networks with such new approach.

Domain adaptation addresses learning tasks where training is performed on data from one domain whereas testing is performed on data belonging to a different but related domain. Assumptions about the relationship between the source and target domains should lead to tractable solutions on the one hand, and be realistic on the other hand. Here we propose a generative domain adaptation model that allows for modelling different assumptions about this relationship, among which is a newly introduced assumption that replaces covariate shift with a possibly more realistic assumption without losing tractability due to the efficient variational inference procedure developed. In addition to the ability to model less restrictive relationships between source and target, modelling can be performed without any target labeled data (unsupervised domain adaptation). We also provide a Rademacher complexity bound of the proposed algorithm. We evaluate the model on the Amazon reviews and the CVC pedestrian detection datasets.

Sum-Product Networks (SPNs) are probabilistic inference machines that admit exact inference in linear time in the size of the network. Existing parameter learning approaches for SPNs are largely based on the maximum likelihood principle and are subject to overfitting compared to more Bayesian approaches. Exact Bayesian posterior inference for SPNs is computationally intractable. Even approximation techniques such as standard variational inference and posterior sampling for SPNs are computationally infeasible even for networks of moderate size due to the large number of local latent variables per instance. In this work, we propose a novel deterministic collapsed variational inference algorithm for SPNs that is computationally efficient, easy to implement and at the same time allows us to incorporate prior information into the optimization formulation. Extensive experiments show a significant improvement in accuracy compared with a maximum likelihood based approach.

The aim of domain adaptation algorithms is to establish a learner, trained on labeled data from a source domain, that can classify samples from a target domain, in which few or no labeled data are available for training. Covariate shift, a primary assumption in several works on domain adaptation, assumes that the labeling functions of source and target domains are identical. We present a domain adaptation algorithm that assumes a relaxed version of covariate shift where the assumption that the labeling functions of the source and target domains are identical holds with a certain probability. Assuming a source deterministic large margin binary classifier, the farther a target instance is from the source decision boundary, the higher the probability that covariate shift holds. In this context, given a target unlabeled sample and no target labeled data, we develop a domain adaptation algorithm that bases its labeling decisions both on the source learner and on the similarities between the target unlabeled instances. The source labeling function decisions associated with probabilistic covariate shift, along with the target similarities are concurrently expressed on a similarity graph. We evaluate our proposed algorithm on a benchmark sentiment analysis (and domain adaptation) dataset, where state-of-the-art adaptation results are achieved. We also derive a lower bound on the performance of the algorithm.

The advantages of soft c-means over its hard and fuzzy versions render it more attractive to use in a wide variety of applications. Its main merit lies in its relatively higher convergence speed, which is more obvious in the presence of huge high dimensional data. This work presents a new approach to accelerate the convergence of the original soft c-means. It is mainly based on an iterative optimization approach and a relaxation technique. Several low and high dimensional datasets are used to evaluate the performance of the proposed approach. Experimental results show up to 70% improvement over the original soft and fuzzy c-means algorithms.