Existing work in counterfactual Learning to Rank (LTR) has focussed on optimizing feature-based models that predict the optimal ranking based on document features. LTR methods based on bandit algorithms often optimize tabular models that memorize the optimal ranking per query. These types of model have their own advantages and disadvantages. Feature-based models provide very robust performance across many queries, including those previously unseen, however, the available features often limit the rankings the model can predict. In contrast, tabular models can converge on any possible ranking through memorization. However, memorization is extremely prone to noise, which makes tabular models reliable only when large numbers of user interactions are available. Can we develop a robust counterfactual LTR method that pursues memorization-based optimization whenever it is safe to do? We introduce the Generalization and Specialization (GENSPEC) algorithm, a robust feature-based counterfactual LTR method that pursues per-query memorization when it is safe to do so. GENSPEC optimizes a single feature-based model for generalization: robust performance across all queries, and many tabular models for specialization: each optimized for high performance on a single query. GENSPEC uses novel relative high-confidence bounds to choose which model to deploy per query. By doing so, GENSPEC enjoys the high performance of successfully specialized tabular models with the robustness of a generalized feature-based model. Our results show that GENSPEC leads to optimal performance on queries with sufficient click data, while having robust behavior on queries with little or noisy data.
Generating tests that can reveal performance issues in large and complex software systems within a reasonable amount of time is a challenging task. On one hand, there are numerous combinations of input data values to explore. On the other hand, we have a limited test budget to execute tests. What makes this task even more difficult is the lack of access to source code and the internal details of these systems. In this paper, we present an automated test generation method called ACTA for black-box performance testing. ACTA is based on active learning, which means that it does not require a large set of historical test data to learn about the performance characteristics of the system under test. Instead, it dynamically chooses the tests to execute using uncertainty sampling. ACTA relies on a conditional variant of generative adversarial networks,and facilitates specifying performance requirements in terms of conditions and generating tests that address those conditions.We have evaluated ACTA on a benchmark web application, and the experimental results indicate that this method is comparable with random testing, and two other machine learning methods,i.e. PerfXRL and DN.
A central goal in designing clinical trials is to find the test that maximizes power (or equivalently minimizes required sample size) for finding a true research hypothesis subject to the constraint of type I error. When there is more than one test, such as in clinical trials with multiple endpoints, the issues of optimal design and optimal policies become more complex. In this paper we address the question of how such optimal tests should be defined and how they can be found. We review different notions of power and how they relate to study goals, and also consider the requirements of type I error control and the nature of the policies. This leads us to formulate the optimal policy problem as an explicit optimization problem with objective and constraints which describe its specific desiderata. We describe a complete solution for deriving optimal policies for two hypotheses, which have desired monotonicity properties, and are computationally simple. For some of the optimization formulations this yields optimal policies that are identical to existing policies, such as Hommel's procedure or the procedure of Bittman et al. (2009), while for others it yields completely novel and more powerful policies than existing ones. We demonstrate the nature of our novel policies and their improved power extensively in simulation and on the APEX study (Cohen et al., 2016).
Optimizing ranking systems based on user interactions is a well-studied problem. State-of-the-art methods for optimizing ranking systems based on user interactions are divided into online approaches - that learn by directly interacting with users - and counterfactual approaches - that learn from historical interactions. Existing online methods are hindered without online interventions and thus should not be applied counterfactually. Conversely, counterfactual methods cannot directly benefit from online interventions. We propose a novel intervention-aware estimator for both counterfactual and online Learning to Rank (LTR). With the introduction of the intervention-aware estimator, we aim to bridge the online/counterfactual LTR division as it is shown to be highly effective in both online and counterfactual scenarios. The estimator corrects for the effect of position bias, trust bias, and item-selection bias by using corrections based on the behavior of the logging policy and on online interventions: changes to the logging policy made during the gathering of click data. Our experimental results, conducted in a semi-synthetic experimental setup, show that, unlike existing counterfactual LTR methods, the intervention-aware estimator can greatly benefit from online interventions.
While existing work in robust deep learning has focused on small pixel-level $\ell_p$ norm-based perturbations, this may not account for perturbations encountered in several real world settings. In many such cases although test data might not be available, broad specifications about the types of perturbations (such as an unknown degree of rotation) may be known. We consider a setup where robustness is expected over an unseen test domain that is not i.i.d. but deviates from the training domain. While this deviation may not be exactly known, its broad characterization is specified a priori, in terms of attributes. We propose an adversarial training approach which learns to generate new samples so as to maximize exposure of the classifier to the attributes-space, without having access to the data from the test domain. Our adversarial training solves a min-max optimization problem, with the inner maximization generating adversarial perturbations, and the outer minimization finding model parameters by optimizing the loss on adversarial perturbations generated from the inner maximization. We demonstrate the applicability of our approach on three types of naturally occurring perturbations -- object-related shifts, geometric transformations, and common image corruptions. Our approach enables deep neural networks to be robust against a wide range of naturally occurring perturbations. We demonstrate the usefulness of the proposed approach by showing the robustness gains of deep neural networks trained using our adversarial training on MNIST, CIFAR-10, and a new variant of the CLEVR dataset.
Machine learning plays a role in many deployed decision systems, often in ways that are difficult or impossible to understand by human stakeholders. Explaining, in a human-understandable way, the relationship between the input and output of machine learning models is essential to the development of trustworthy machine-learning-based systems. A burgeoning body of research seeks to define the goals and methods of explainability in machine learning. In this paper, we seek to review and categorize research on counterfactual explanations, a specific class of explanation that provides a link between what could have happened had input to a model been changed in a particular way. Modern approaches to counterfactual explainability in machine learning draw connections to the established legal doctrine in many countries, making them appealing to fielded systems in high-impact areas such as finance and healthcare. Thus, we design a rubric with desirable properties of counterfactual explanation algorithms and comprehensively evaluate all currently-proposed algorithms against that rubric. Our rubric provides easy comparison and comprehension of the advantages and disadvantages of different approaches and serves as an introduction to major research themes in this field. We also identify gaps and discuss promising research directions in the space of counterfactual explainability.
Active learning from demonstration allows a robot to query a human for specific types of input to achieve efficient learning. Existing work has explored a variety of active query strategies; however, to our knowledge, none of these strategies directly minimize the performance risk of the policy the robot is learning. Utilizing recent advances in performance bounds for inverse reinforcement learning, we propose a risk-aware active inverse reinforcement learning algorithm that focuses active queries on areas of the state space with the potential for large generalization error. We show that risk-aware active learning outperforms standard active IRL approaches on gridworld, simulated driving, and table setting tasks, while also providing a performance-based stopping criterion that allows a robot to know when it has received enough demonstrations to safely perform a task.
Deep reinforcement learning suggests the promise of fully automated learning of robotic control policies that directly map sensory inputs to low-level actions. However, applying deep reinforcement learning methods on real-world robots is exceptionally difficult, due both to the sample complexity and, just as importantly, the sensitivity of such methods to hyperparameters. While hyperparameter tuning can be performed in parallel in simulated domains, it is usually impractical to tune hyperparameters directly on real-world robotic platforms, especially legged platforms like quadrupedal robots that can be damaged through extensive trial-and-error learning. In this paper, we develop a stable variant of the soft actor-critic deep reinforcement learning algorithm that requires minimal hyperparameter tuning, while also requiring only a modest number of trials to learn multilayer neural network policies. This algorithm is based on the framework of maximum entropy reinforcement learning, and automatically trades off exploration against exploitation by dynamically and automatically tuning a temperature parameter that determines the stochasticity of the policy. We show that this method achieves state-of-the-art performance on four standard benchmark environments. We then demonstrate that it can be used to learn quadrupedal locomotion gaits on a real-world Minitaur robot, learning to walk from scratch directly in the real world in two hours of training.
The reinforcement learning community has made great strides in designing algorithms capable of exceeding human performance on specific tasks. These algorithms are mostly trained one task at the time, each new task requiring to train a brand new agent instance. This means the learning algorithm is general, but each solution is not; each agent can only solve the one task it was trained on. In this work, we study the problem of learning to master not one but multiple sequential-decision tasks at once. A general issue in multi-task learning is that a balance must be found between the needs of multiple tasks competing for the limited resources of a single learning system. Many learning algorithms can get distracted by certain tasks in the set of tasks to solve. Such tasks appear more salient to the learning process, for instance because of the density or magnitude of the in-task rewards. This causes the algorithm to focus on those salient tasks at the expense of generality. We propose to automatically adapt the contribution of each task to the agent's updates, so that all tasks have a similar impact on the learning dynamics. This resulted in state of the art performance on learning to play all games in a set of 57 diverse Atari games. Excitingly, our method learned a single trained policy - with a single set of weights - that exceeds median human performance. To our knowledge, this was the first time a single agent surpassed human-level performance on this multi-task domain. The same approach also demonstrated state of the art performance on a set of 30 tasks in the 3D reinforcement learning platform DeepMind Lab.
Most policy search algorithms require thousands of training episodes to find an effective policy, which is often infeasible with a physical robot. This survey article focuses on the extreme other end of the spectrum: how can a robot adapt with only a handful of trials (a dozen) and a few minutes? By analogy with the word "big-data", we refer to this challenge as "micro-data reinforcement learning". We show that a first strategy is to leverage prior knowledge on the policy structure (e.g., dynamic movement primitives), on the policy parameters (e.g., demonstrations), or on the dynamics (e.g., simulators). A second strategy is to create data-driven surrogate models of the expected reward (e.g., Bayesian optimization) or the dynamical model (e.g., model-based policy search), so that the policy optimizer queries the model instead of the real system. Overall, all successful micro-data algorithms combine these two strategies by varying the kind of model and prior knowledge. The current scientific challenges essentially revolve around scaling up to complex robots (e.g., humanoids), designing generic priors, and optimizing the computing time.
In many real-world settings, a team of agents must coordinate their behaviour while acting in a decentralised way. At the same time, it is often possible to train the agents in a centralised fashion in a simulated or laboratory setting, where global state information is available and communication constraints are lifted. Learning joint action-values conditioned on extra state information is an attractive way to exploit centralised learning, but the best strategy for then extracting decentralised policies is unclear. Our solution is QMIX, a novel value-based method that can train decentralised policies in a centralised end-to-end fashion. QMIX employs a network that estimates joint action-values as a complex non-linear combination of per-agent values that condition only on local observations. We structurally enforce that the joint-action value is monotonic in the per-agent values, which allows tractable maximisation of the joint action-value in off-policy learning, and guarantees consistency between the centralised and decentralised policies. We evaluate QMIX on a challenging set of StarCraft II micromanagement tasks, and show that QMIX significantly outperforms existing value-based multi-agent reinforcement learning methods.