Modified backpropagation methods are a popular group of attribution methods. We analyse the most prominent methods: Deep Taylor Decomposition, Layer-wise Relevance Propagation, Excitation BP, PatternAttribution, Deconv, and Guided BP. We found empirically that the explanations of the mentioned modified BP methods are independent of the parameters of later layers and show that the $z^+$ rule used by multiple methods converges to a rank-1 matrix. This can explain well why the actual network's decision is ignored. We also develop a new metric cosine similarity convergence (CSC) to directly quantify the convergence of the modified BP methods to a rank-1 matrix. Our conclusion is that many modified BP methods do not explain the predictions of deep neural networks faithfully.

Though used extensively, the concept and process of machine learning (ML) personalization have generally received little attention from academics, practitioners, and the general public. We describe the ML approach as relying on the metaphor of the person as a feature vector and contrast this with humanistic views of the person. In light of the recent calls by the IEEE to consider the effects of ML on human well-being, we ask whether ML personalization can be reconciled with these humanistic views of the person, which highlight the importance of moral and social identity. As human behavior increasingly becomes digitized, analyzed, and predicted, to what extent do our subsequent decisions about what to choose, buy, or do, made both by us and others, reflect who we are as persons? This paper first explicates the term personalization by considering ML personalization and highlights its relation to humanistic conceptions of the person, then proposes several dimensions for evaluating the degree of personalization of ML personalized scores. By doing so, we hope to contribute to current debate on the issues of algorithmic bias, transparency, and fairness in machine learning.

Many dynamical systems exhibit similar structure, as often captured by hand-designed simplified models that can be used for analysis and control. We develop a method for learning to correspond pairs of dynamical systems via a learned latent dynamical system. Given trajectory data from two dynamical systems, we learn a shared latent state space and a shared latent dynamics model, along with an encoder-decoder pair for each of the original systems. With the learned correspondences in place, we can use a simulation of one system to produce an imagined motion of its counterpart. We can also simulate in the learned latent dynamics and synthesize the motions of both corresponding systems, as a form of bisimulation. We demonstrate the approach using pairs of controlled bipedal walkers, as well as by pairing a walker with a controlled pendulum.

Given a set of source data with pre-trained classification models, how can we fast and accurately select the most useful source data to improve the performance of a target task? We address the problem of measuring transferability for heterogeneous domains, where the source and the target data have different feature spaces and distributions. We propose Transmeter, a novel method to efficiently and accurately measure transferability of two datasets. Transmeter utilizes a pre-trained source classifier and a reconstruction loss to increase its efficiency and performance. Furthermore, Transmeter uses feature transformation layers, label-wise discriminators, and a mean distance loss to learn common representations for source and target domains. As a result, Transmeter and its variant give the most accurate performance in measuring transferability, while giving comparable running times compared to those of competitors.

We address two central notions of fairness in the literature of planning on nondeterministic fully observable domains. The first, which we call stochastic fairness, is classical, and assumes an environment which operates probabilistically using possibly unknown probabilities. The second, which is language-theoretic, assumes that if an action is taken from a given state infinitely often then all its possible outcomes should appear infinitely often (we call this state-action fairness). While the two notions coincide for standard reachability goals, they diverge for temporally extended goals. This important difference has been overlooked in the planning literature, and we argue has led to confusion in a number of published algorithms which use reductions that were stated for state-action fairness, for which they are incorrect, while being correct for stochastic fairness. We remedy this and provide an optimal sound and complete algorithm for solving state-action fair planning for LTL/LTLf goals, as well as a correct proof of the lower bound of the goal-complexity (our proof is general enough that it provides new proofs also for the no-fairness and stochastic-fairness cases). Overall, we show that stochastic fairness is better behaved than state-action fairness.

Olivier Delalleau * 1, Maxim Peter *, Eloi Alonso, Adrien Logut Ubisoft La Forge Abstract While most current research in Reinforcement Learning (RL) focuses on improving the performance of the algorithms in controlled environments, the use of RL under constraints like those met in the video game industry is rarely studied. Operating under such constraints, we propose Hybrid SAC, an extension of the Soft Actor-Critic algorithm able to handle discrete, continuous and parameterized actions in a principled way. We show that Hybrid SAC can successfully solve a high-speed driving task in one of our games, and is competitive with the state-of-the-art on parameterized actions benchmark tasks. We also explore the impact of using normalizing flows to enrich the expressiveness of the policy at minimal computational cost, and identify a potential undesired effect of SAC when used with normalizing flows, that may be addressed by optimizing a different objective. Introduction Reinforcement Learning (RL) applications in video games have recently seen massive advances coming from the research community, with agents trained to play Atari games from pixels (Mnih et al. 2015) or to be competitive with the best players in the world in complicated imperfect information games like DOT A 2 (OpenAI 2018) or StarCraft II (Vinyals et al. 2019a; 2019b). These systems have comparatively seen little use within the video game industry, and we believe lack of accessibility to be a major reason behind this. Indeed, really impressive results like those cited above are produced by large research groups with computational resources well beyond what is typically available within video game studios. Our contributions are geared towards industry practitioners, by sharing experiments and practical advice for using RL with a different set of constraints than those met in the research community.

Learning robotic manipulation tasks using reinforcement learning with sparse rewards is currently impractical due to the outrageous data requirements. Many practical tasks require manipulation of multiple objects, and the complexity of such tasks increases with the number of objects. Learning from a curriculum of increasingly complex tasks appears to be a natural solution, but unfortunately, does not work for many scenarios. We hypothesize that the inability of the state-of-the-art algorithms to effectively utilize a task curriculum stems from the absence of inductive biases for transferring knowledge from simpler to complex tasks. We show that graph-based relational architectures overcome this limitation and enable learning of complex tasks when provided with a simple curriculum of tasks with increasing numbers of objects. We demonstrate the utility of our framework on a simulated block stacking task. Starting from scratch, our agent learns to stack six blocks into a tower. Despite using step-wise sparse rewards, our method is orders of magnitude more data-efficient and outperforms the existing state-of-the-art method that utilizes human demonstrations. Furthermore, the learned policy exhibits zero-shot generalization, successfully stacking blocks into taller towers and previously unseen configurations such as pyramids, without any further training.

Minimum Routing Cost Clustered Tree Problem (CluMRCT) is applied in various fields in both theory and application. Because the CluMRCT is NP-Hard, the approximate approaches are suitable to find the solution for this problem. Recently, Multifactorial Evolutionary Algorithm (MFEA) has emerged as one of the most efficient approximation algorithms to deal with many different kinds of problems. Therefore, this paper studies to apply MFEA for solving CluMRCT problems. In the proposed MFEA, we focus on crossover and mutation operators which create a valid solution of CluMRCT problem in two levels: first level constructs spanning trees for graphs in clusters while the second level builds a spanning tree for connecting among clusters. To reduce the consuming resources, we will also introduce a new method of calculating the cost of CluMRCT solution. The proposed algorithm is experimented on numerous types of datasets. The experimental results demonstrate the effectiveness of the proposed algorithm, partially on large instances

Abstract--Backtracking search algorithms are often used to solve the Constraint Satisfaction Problem (CSP). The efficiency of backtracking search depends greatly on the variable ordering heuristics. Currently, the most commonly used heuristics are handcrafted based on expert knowledge. In this paper, we propose a deep reinforcement learning based approach to automatically discover new variable ordering heuristics that are better adapted for a given class of CSP instances. We show that directly optimizing the search cost is hard for bootstrapping, and propose to optimize the expected cost of reaching a leaf node in the search tree. T o capture the complex relations among the variables and constraints, we design a representation scheme based on Graph Neural Network that can process CSP instances with different sizes and constraint arities. Experimental results on random CSP instances show that the learned policies outperform classical handcrafted heuristics in terms of minimizing the search tree size, and can effectively generalize to instances that are larger than those used in training. Constraint Satisfaction Problem (CSP) is one of the most widely studied problems in computer science and artificial intelligence. It provides a common framework for modeling and solving combinatorial problems in many application domains, such as planning and scheduling [1], [2], vehicle routing [3], [4], graph problems [5], [6], and computational biology [7], [8]. A CSP instance involves a set of variables and constraints. T o solve it, one needs to find a value assignment for all variables such that all constraints are satisfied, or prove such assignment does not exist. Despite its ubiquitous applications, unfortunately, CSP is well known to be NPcomplete in general [9]. T o solve CSP efficiently, backtracking search algorithms are often employed, which are exact algorithms with the guarantee that a solution will be found if one exists.

Artificial Intelligence can help detect one of the most common forms of blood cancer - acute myeloid leukaemia (AML) - with high reliability, new research has found. Their approach, based on the analysis of the gene activity of cells found in the blood, could support conventional diagnostics and possibly accelerate the beginning of therapy, said the study published in the journal iScience. In the early stages the symptoms of AML can resemble those of a bad cold. However, AML is a life-threatening disease that should be treated as quickly as possible. "With a blood test, as it seems possible on the basis of our study, it is conceivable that the family doctor would already clarify a suspicion of AML," said Joachim Schultze, a research group leader at the German Center for Neurodegenerative Diseases (DZNE).