The similarity assumption in Case-Based Reasoning (similar problems have similar solutions) has been questioned by several researchers. If knowledge about the adaptability of solutions is available, it can be exploited in order to guide retrieval. Several approaches have been proposed in this context, often assuming a similarity or cost measure defined over the solution space. In this paper, we propose a novel approach where the adaptability of the solutions is captured inside a metric Markov Random Field (MRF). Each case is represented as a node in the MRF, and edges connect cases whose solutions are close in the solution space. States of the nodes represent the adaptability effort with respect to the query. Potentals are defined to enforce connected nodes to share the same state; this models the fact that cases having similar solutions should have the same adaptability effort with respect to the query. The main goal is to enlarge the set of potentially adaptable cases that are retrieved (the recall) without significantly sacrificing the precision of retrieval. We will report on some experiments concerning a retrieval architecture where a simple kNN retrieval is followed by a further retrieval step based on MRF inference.

Learning from Observation (LfO) is highly useful for modeling behaviors through nonintrusive observation of some actor's performance. However, an actor's performance is often influenced by unobservable internal influences, such as emotions, agendas, and memory of past events. Therefore, new techniques are needed to infer the structure of these influences and their effect on an actor's decisions. In this paper, we propose a novel approach called Memory Composition Learning (MCL) for capturing one internal influence: memory of past events. We hypothesize that memory influences on a behavior can be modeled through parameterized memory features that can be learned from observation of traces of an actor's behavior; these memory features can then be presented as additional input to a performance modeling application. We demonstrate the efficacy of our approach in a simulated vacuum cleaner domain and show that hidden memory influences can be detected, modeled, and then used to improve machine learning performance.

Conditional independence structures describe independencies of one set of variables from another set of variables conditioned upon a third set of variables. These structures are invaluable means for compact representations of knowledge because independencies can be exploited for useful factorizations. Conditional independence structures appear in different disguise in various areas of knowledge representation, be it the conditional independence of sets of random variables in probabilistic graphical models such as Bayesian networks or as conditional functions related to belief revision, or as in- dependencies induced by (embedded) multivalued dependencies in data bases. This paper investigates conditional independencies for Boolean functions using Fourier analysis. We define three notions of independence based on the notion of influence of a variable on a function and draw connections to multivalued dependencies.

Machine learning and data mining techniques have been used extensively in order to detect credit card frauds. However, most studies consider credit card transactions as isolated events and not as a sequence of transactions. In this article, we model a sequence of credit card transactions from three different perspectives, namely (i) does the sequence contain a Fraud? (ii) Is the sequence obtained by fixing the card-holder or the payment terminal? (iii) Is it a sequence of spent amount or of elapsed time between the current and previous transactions? Combinations of the three binary perspectives give eight sets of sequences from the (training) set of transactions. Each one of these sets is modelled with a Hidden Markov Model (HMM). Each HMM associates a likelihood to a transaction given its sequence of previous transactions. These likelihoods are used as additional features in a Random Forest classifier for fraud detection. This multiple perspectives HMM-based approach enables an automatic feature engineering in order to model the sequential properties of the dataset with respect to the classification task. This strategy allows for a 15% increase in the precision-recall AUC compared to the state of the art feature engineering strategy for credit card fraud detection.

Rewards in Markov Decision Processes (MDP) define the behavior of the model. Without a clear interpretation of what the reward function is and is not capturing, one cannot trust their model nor diagnose when the model is giving incorrect recommendations. Increasing complexity of state-of-the-art models used to represent the reward function and model-free methods that attempt to avoid representing this function make trusting the model much more difficult. We map these reward functions onto a standard classification problem where we can explain what factors the model considers in making decisions in local and global contexts and quantify whether the fit of the reward function is likely to be good for explaining the behavior of the model. We evaluate our proof-of-concept on both the standard version and a modified version of the Object World domain to add more nonlinearity.

In this work is proposed a method for Hierarchical Classification, which takes advantage of the hierarchical structure to influence the prediction of local classifiers with their neighbors. To achieve this, two strategies are combined. The first is to represent the hierarchical structure as a Bayesian network, and the second is to build chained classifiers that feed the Bayesian network as local classifiers. The proposed method was tested in several datasets of functional genomics, which consist of tree-structured hierarchies. The results of several variants of the proposed method are compared to the standard methods, Flat and Top-Down, as well as with a start of the art technique, showing superior performance under several metrics.

Two different dynamic models for belief change during evidence monitoring were evaluated: Markov and quantum. They were empirically tested with an experiment in which participants monitored evidence for an initial period of time, made a probability rating, then monitored more evidence, before making a second rating. The models were qualitatively tested by manipulating the time intervals in a manner that provided a test for interference effects of the first rating on the second. The Markov model predicted no interference whereas the quantum model predicted interference. A quantitative comparison of the two models was also carried out using a generalization criterion method: the parameters were fit to data from one set of time intervals, and then these same parameters were used to predict data from another set of time intervals. The results indicated that some features of both Markov and quantum models are needed to accurately account for the results.

Humans achieve efficient learning by relying on prior knowledge about the structure of naturally occurring tasks. There has been considerable interest in designing reinforcement learning algorithms with similar properties. This includes several proposals to learn the learning algorithm itself, an idea also referred to as meta learning. One formal interpretation of this idea is in terms of a partially observable multi-task reinforcement learning problem in which information about the task is hidden from the agent. Although agents that solve partially observable environments can be trained from rewards alone, shaping an agent's memory with additional supervision has been shown to boost learning efficiency. It is thus natural to ask what kind of supervision, if any, facilitates meta-learning. Here we explore several choices and develop an architecture that separates learning of the belief about the unknown task from learning of the policy, and that can be used effectively with privileged information about the task during training. We show that this approach can be very effective at solving standard meta-RL environments, as well as a complex continuous control environment in which a simulated robot has to execute various movement sequences.

Bayesian neural network (BNN) priors are defined in parameter space, making it hard to encode prior knowledge expressed in function space. We formulate a prior that incorporates functional constraints about what the output can or cannot be in regions of the input space. Output-Constrained BNNs (OC-BNN) represent an interpretable approach of enforcing a range of constraints, fully consistent with the Bayesian framework and amenable to black-box inference. We demonstrate how OC-BNNs improve model robustness and prevent the prediction of infeasible outputs in two real-world applications of healthcare and robotics.

Motivated by the study of $Q$-learning algorithms in reinforcement learning, we study a class of stochastic approximation procedures based on operators that satisfy monotonicity and quasi-contractivity conditions with respect to an underlying cone. We prove a general sandwich relation on the iterate error at each time, and use it to derive non-asymptotic bounds on the error in terms of a cone-induced gauge norm. These results are derived within a deterministic framework, requiring no assumptions on the noise. We illustrate these general bounds in application to synchronous $Q$-learning for discounted Markov decision processes with discrete state-action spaces, in particular by deriving non-asymptotic bounds on the $\ell_\infty$-norm for a range of stepsizes. These results are the sharpest known to date, and we show via simulation that the dependence of our bounds cannot be improved in a worst-case sense. These results show that relative to a model-based $Q$-iteration, the $\ell_\infty$-based sample complexity of $Q$-learning is suboptimal in terms of the discount factor $\gamma$.