Probabilistic inverse reinforcement learning in unknown environments

We consider the problem of learning by demonstration from agents acting in unknown stochastic Markov environments or games. Our aim is to estimate agent preferences in order to construct improved policies for the same task that the agents are trying to solve. To do so, we extend previous probabilistic approaches for inverse reinforcement learning in known MDPs to the case of unknown dynamics or opponents. We do this by deriving two simplified probabilistic models of the demonstrator's policy and utility. For tractability, we use maximum a posteriori estimation rather than full Bayesian inference. Under a flat prior, this results in a convex optimisation problem. We find that the resulting algorithms are highly competitive against a variety of other methods for inverse reinforcement learning that do have knowledge of the dynamics.

Mean-Field methods for Structured Deep-Learning in Computer Vision

Pierre Bruno Baqué

In recent years, Machine Learning based Computer Vision techniques made impressive progress. These algorithms proved particularly efficient for image classification or detection of isolated objects. From a probabilistic perspective, these methods can predict marginals, over single or multiple variables, independently, with high accuracy. However, in many tasks of practical interest, we need to predict jointly several correlated variables. Practical applications include people detection in crowded scenes, image segmentation, surface reconstruction, 3D pose estimation and others. A large part of the research effort in today's computer-vision community aims at finding task-specific solutions to these problems, while leveraging the power of Deep-Learning based classifiers. In this thesis, we present our journey towards a generic and practical solution based on mean-field (MF) inference. Mean-field is a Statistical Physics-inspired method which has long been used in Computer-Vision as a variational approximation to posterior distributions over complex Conditional Random Fields. Standard mean-field optimization is based on coordinate descent and in many situations can be impractical. We therefore propose a novel proximal gradient-based approach to optimizing the variational objective. It is naturally parallelizable and easy to implement. We prove its convergence, and then demonstrate that, in practice, it yields faster convergence and often finds better optima than more traditional mean-field optimization techniques. Then, we show that we can replace the fully factorized distribution of mean-field by a weighted mixture of such distributions, that similarly minimizes the KL-Divergence to the true posterior. Our extension of the clamping method proposed in previous works allows us to both produce a more descriptive approximation of the true posterior and, inspired by the diverse MAP paradigms, fit a mixture of mean-field approximations. We demonstrate that this positively impacts real-world algorithms that initially relied on mean-fields. One of the important properties of the mean-field inference algorithms is that the closed-form updates are fully differentiable operations. This naturally allows to do parameter learning by simply unrolling multiple iterations of the updates, the so-called back-mean-field algorithm. We derive a novel and efficient structured learning method for multi-modal posterior distribution based on the Multi-Modal Mean-Field approximation, which can be seamlessly combined to modern gradient-based learning methods such as CNNs. Finally, we explore in more details the specific problem of structured learning and prediction for multiple-people detection in crowded scenes. We then present a mean-field based structured deep-learning detection algorithm that provides state of the art results on this dataset.

EPFL2018

Surprise-based model estimation in reinforcement learning: algorithms and brain signatures

Vasiliki Liakoni

Learning how to act and adapting to unexpected changes are remarkable capabilities of humans and other animals. In the absence of a direct recipe to follow in life, behaviour is often guided by rewarding and by surprising events. A positive or a negative outcome influences the tendency to repeat some actions, and a sudden unexpected event signals the possible need to act differently or to update one's view about the world. Advances in computational, behavioral, and cognitive neuroscience have indicated that animals employ multiple strategies to learn from interaction. However, our understanding of learning strategies and how they are combined is largely restricted. The main goal of this thesis is to study the use of surprise by ever-adapting biological agents, its contributions to reward-based learning, and its manifestation in the human brain. We first study surprise from a theoretical perspective. In a probabilistic model of changing environments, we show that exact and approximate Bayesian inference give rise to a trade-off between forgetting old observations and integrating them with new ones, modulated by a naturally emerging surprise measure. We develop novel surprised-based algorithms that can adapt in the face of abrupt changes and accurately estimate the model of the world, and that could potentially be implemented in the brain. Next, we focus on the contributions of surprise-based model estimation to reinforcement learning. We couple one of our adaptive algorithms as well as simpler non-adaptive methods with reinforcement learning agents and evaluate their performance on environments exhibiting different characteristics. Abrupt changes that directly affect the agent's policy call for surprise-based adaptation, in order to achieve higher performance. Often, however, the agent does not need to invest in maintaining an accurate model of the environment to obtain high reward levels. More specifically, in stochastic environments or in environments with distal changes, simpler methods, equipped with exploration capacity, perform equally well compared to more elaborate methods. Finally, we turn to human learning behaviour and brain signals of surprise- and reward-based learning. We design a novel sequential decision making task of multiple steps where strategic use of surprising events allows us to dissociate fMRI brain correlates of reward learning and model estimation. We show that Bayesian inference on this task leads to the same surprise measure we found earlier, where the trade-off is now between ignoring new observations and integrating them with the old belief, and we develop reinforcement learning algorithms that perform outlier detection via this surprise-modulated trade-off. At the level of behaviour we find evidence for a model-free policy learning architecture, with potential influences from a model estimation system. At the level of brain responses we identify signatures of both reward- and model estimation signals, supporting the existence of multiple parallel learning systems in the brain. This thesis presents a comparative analysis of surprise-based model estimation methods in theory and simulations, provides insights in the type of approximations that biological agents may adopt, and identifies signatures of model estimation in the human brain. Our results may aid future work aiming at building efficient adaptive agents and at understanding the learning algorithms and the surprise measures implemented in the brain.

EPFL2021

Learning to Reduce Annotation Load

Ksenia Konyushkova

Modern machine learning methods and their applications in computer vision are known to crave for large amounts of training data to reach their full potential. Because training data is mostly obtained through humans who manually label samples, it induces a significant cost. Therefore, the problem of reducing the annotation load is of great importance for the success of machine learning methods. We study the problem of reducing the annotation load from two viewpoints, by answering the questions âWhat to annotate?â and âHow to annotate?â. The question âWhat?â addresses the selection of a small portion of the data that would be sufficient to train an accurate model. The question âHow? focuses on minimising the effort of labelling each datapoint. The question âWhat to annotate?â becomes particularly compelling if we can select data to be annotated in an iterative and adaptive way, a setting known as active learning (AL). The key challenge in AL is to identify the datapoints that are the most informative for the model at a given stage. We propose several techniques to address this challenge. Firstly, we consider the problem of segmenting natural images and image volumes. We take advantage of image priors, such as smoothness of objects of interest, and use them in a novel form of geometric uncertainty. Using this, we design an AL technique to efficiently annotate data that is tailored to segmentation applications. Next, we notice that no single manually-designed strategy outperforms others in every application and that often the burden of designing new strategies outweighs the benefits of AL. To overcome this problem we suggest learning an AL strategy from data by formulating the AL problem as a regression task that predicts the reduction in the generalisation error achieved by labelling each datapoint. This enables us to learn AL strategies from simulated data and to transfer them to new datasets. Finally, we turn towards non-myopic data-driven AL strategies. To this end, we formulate the AL problem as a Markov decision process and find the best selection policy using reinforcement learning. We design the decision process such that the policy can be learnt for any ML model and transferred to diverse application domains. Effectively addressing the question âHow to annotate?â is of no less importance as large cost savings can be achieved by labelling each datapoint more efficiently. This can be done with intelligent interfaces that interact with a human annotator. We make two contributions towards answering the question âHow?â. Firstly, we propose an efficient technique to annotate 3D image volumes for image segmentation. Annotating data in 3D is cumbersome and an obvious way to facilitate it is to select a subset of the data lying on a 2D plane. To find the optimal plane (i.e. the one containing the most informative datapoints) we design a branch-and-bound algorithm that quickly eliminates hypotheses about the optimal projection. Secondly, we propose an intelligent data annotation method to train object detectors. Instead of always asking the human annotator to draw bounding boxes in images, we detect automatically in which cases we can rely on the current detector and verify its proposal.

EPFL2019