A lot of real-world applications could be modeled as the Mixed-Integer Non-Linear Programming (MINLP) problems, and some prominent examples include portfolio optimization, resource allocation, image classification, as well as path planning. Actually, most of the models for these applications are non-convex and always involve some conflicting objectives. Hence, the Multi-Objective Evolutionary Algorithm (MOEA), which does not require the gradient information and is efficient at dealing with the multi-objective optimization problems, is adopted frequently for these problems. In this work, we discuss the coding scheme for MOEA in MINLP, and the major discussion focuses on the constrained portfolio optimization problem, which is a classic financial problem and could be naturally modeled as MINLP. As a result, the challenge, faced by a direct coding scheme for MOEA in MINLP, is pointed out that the searching in multiple search spaces is very complicated. Thus, a Compressed Coding Scheme (CCS), which converts an MINLP problem into a continuous problem, is proposed to address this challenge. The analyses and experiments on 20 portfolio benchmark instances, of which the number of available assets ranging from 31 to 2235, consistently indicate that CCS is not only efficient but also robust for dealing with the constrained multi-objective portfolio optimization.

This paper presents an efficient neural network model to generate robotic grasps with high resolution images. The proposed model uses fully convolution neural network to generate robotic grasps for each pixel using 400 $\times$ 400 high resolution RGB-D images. It first down-sample the images to get features and then up-sample those features to the original size of the input as well as combines local and global features from different feature maps. Compared to other regression or classification methods for detecting robotic grasps, our method looks more like the segmentation methods which solves the problem through pixel-wise ways. We use Cornell Grasp Dataset to train and evaluate the model and get high accuracy about 94.42% for image-wise and 91.02% for object-wise and fast prediction time about 8ms. We also demonstrate that without training on the multiple objects dataset, our model can directly output robotic grasps candidates for different objects because of the pixel wise implementation.

This paper takes a step towards theoretical analysis of the relationship between word embeddings and context embeddings in models such as word2vec. We start from basic probabilistic assumptions on the nature of word vectors, context vectors, and text generation. These assumptions are supported either empirically or theoretically by the existing literature. Next, we show that under these assumptions the widely-used word-word PMI matrix is approximately a random symmetric Gaussian ensemble. This, in turn, implies that context vectors are reflections of word vectors in approximately half the dimensions. As a direct application of our result, we suggest a theoretically grounded way of tying weights in the SGNS model.

Abstract--This paper presents an analysis of pre-trained models to recognize handwritten Devanagari alphabets using transfer learning for Deep Convolution Neural Network (DCNN). This research implements AlexNet, DenseNet, Vgg, and Inception ConvNet as a fixed feature extractor. We implemented 15 epochs for each of AlexNet, DenseNet 121, DenseNet 201, Vgg 11, Vgg 16, Vgg 19, and Inception V3. Results show that Inception V3 performs better in terms of accuracy achieving 99% accuracy with average epoch time 16.3 minutes while AlexNet performs fastest with 2.2 minutes per epoch and achieving 98% accuracy. Handwriting identification is one of the challenging research domains in computer vision. Handwriting character identification systems are useful for bank signatures, postal code recognition, and bank cheques, etc.

Given a set of Datalog rules, facts, and a query, answers to the query can be inferred bottom-up starting from the facts or top-down starting from the query. For efficiency, top-down evaluation is extended with memoization of inferred facts, and bottom-up evaluation is performed after transformations to make rules driven by the demand from the query. Prior work has shown their precise complexity analysis and relationships. However, when Datalog is extended with even stratified negation, which has a simple and universally accepted semantics, transformations to make rules demand-driven may result in non-stratified negation, which has had many complex semantics and evaluation methods. This paper presents (1) a simple extension to demand transformation, a transformation to make rules demand-driven for Datalog without negation, to support stratified negation, and (2) a simple extension to an optimal bottom-up evaluation method for Datalog with stratified negation, to handle non-stratified negation in the resulting rules. We show that the method provides precise complexity guarantees. It is also optimal in that only facts needed for top-down evaluation of the query are inferred and each firing of a rule to infer such a fact takes worst-case constant time. We extend the precise relationship between top-down evaluation and demand-driven bottom-up evaluation to Datalog with stratified negation. Finally, we show experimental results for performance, as well as applications to previously challenging examples.

We evaluate the folk wisdom that algorithms trained on data produced by biased human decision-makers necessarily reflect this bias. We consider a setting where training labels are only generated if a biased decision-maker takes a particular action, and so bias arises due to selection into the training data. In our baseline model, the more biased the decision-maker is toward a group, the more the algorithm favors that group. We refer to this phenomenon as "algorithmic affirmative action." We then clarify the conditions that give rise to algorithmic affirmative action. Whether a prediction algorithm reverses or inherits bias depends critically on how the decision-maker affects the training data as well as the label used in training. We illustrate our main theoretical results in a simulation study applied to the New York City Stop, Question and Frisk dataset.

Quantum Process Tomography (QPT) methods aim at identifying, i.e. estimating, a given quantum process. QPT is a major quantum information processing tool, since it especially allows one to characterize the actual behavior of quantum gates, which are the building blocks of quantum computers. However, usual QPT procedures are complicated, since they set several constraints on the quantum states used as inputs of the process to be characterized. In this paper, we extend QPT so as to avoid two such constraints. On the one hand, usual QPT methods requires one to know, hence to precisely control (i.e. prepare), the specific quantum states used as inputs of the considered quantum process, which is cumbersome. We therefore propose a Blind, or unsupervised, extension of QPT (i.e. BQPT), which means that this approach uses input quantum states whose values are unknown and arbitrary, except that they are requested to meet some general known properties (and this approach exploits the output states of the considered quantum process). On the other hand, usual QPT methods require one to be able to prepare many copies of the same (known) input state, which is constraining. On the contrary, we propose "single-preparation methods", i.e. methods which can operate with only one instance of each considered input state. These two new concepts are here illustrated with practical BQPT methods which are numerically validated, in the case when: i) random pure states are used as inputs and their required properties are especially related to the statistical independence of the random variables that define them, ii) the considered quantum process is based on cylindrical-symmetry Heisenberg spin coupling. These concepts may be extended to a much wider class of processes and to BQPT methods based on other input quantum state properties.

Neural networks are popular models for regression. They are often trained via back-propagation to find a value of the weights that correctly predicts the observed data. Although back-propagation has shown good performance in many applications, it cannot easily output an estimate of the uncertainty in the predictions made. Measuring this uncertainty in the predictions of machine learning models is a critical aspect with important applications. Uncertainty estimates can be obtained by following a Bayesian approach in which a posterior distribution of the model parameters is computed. The posterior distribution summarizes which parameter values are compatible with the data. Typically,this posterior distribution is intractable and has to be approximated. Several approaches have been considered for solving this problem. We propose here a general method for approximate Bayesian inference based on minimizing{\alpha}-divergences which allows for flexible approximate distributions. The method is evaluated in the context of Bayesian neural networks for regression on extensive experiments. The results show that it often gives better performance in terms of the test log-likelihood and sometimes in terms of the squared error.

The aim of my Ph.D. thesis concerns Reasoning in Highly Reactive Environments. As reasoning in highly reactive environments, we identify the setting in which a knowledge-based agent, with given goals, is deployed in an environment subject to repeated, sudden and possibly unknown changes. This is for instance the typical setting in which, e.g., artificial agents for video-games (the so called "bots"), cleaning robots, bomb clearing robots, and so on are deployed. In all these settings one can follow the classical approach in which the operations of the agent are distinguished in "sensing" the environment with proper interface devices, "thinking", and then behaving accordingly using proper actuators. In order to operate in an highly reactive environment, an artificial agent needs to be: 1. Responsive -> The agent must be able to react repeatedly and in a reasonable amount of time; 2. Elastic -> The agent must stay reactive also under varying workload; 3. Resilient -> The agent must stay responsive also in case of internal failure or failure of one of the programmed actions in the environment. Nowadays, thanks to new technologies in the field of Artificial Intelligence, it is already technically possible to create AI agents that are able to operate in reactive environments. Nevertheless, several issues stay unsolved, and are subject of ongoing research.

Boolean networks are conventionally used to represent and simulate gene regulatory networks. In the analysis of the dynamic of a Boolean network, the attractors are the objects of a special attention. In this work, we propose a novel approach based on Answer Set Programming (ASP) to express Boolean networks and simulate the dynamics of such networks. Our work focuses on the identification of the attractors, it relies on the exhaustive enumeration of all the attractors of synchronous and asynchronous Boolean networks. We applied and evaluated the proposed approach on real biological networks, and the obtained results indicate that this novel approach is promising.