Developing new packaged foods and consumer goods can take a couple years as companies research, prototype and test products. In a society that runs on social media, however, people expect to see trends land on store shelves much more quickly. Founded in 2018, Ai Palette uses machine learning to help companies spot trends in real time and get them retail-ready, often within a few months. The startup, whose clients include Danone, Kellogg's, Cargill and Dole, announced today it has raised an oversubscribed $4.4 million Series A co-led by pi Ventures and Exfinity Venture Partners. Both will join Ai Palette's board.

Recent works have demonstrated the remarkable successes of deep reinforcement learning in various domains including finance, medicine, healthcare, video games, robotics, and computer vision. In this work, we provide a detailed review of recent and state-of-the-art research advances of deep reinforcement learning in computer vision. We start with comprehending the theories of deep learning, reinforcement learning, and deep reinforcement learning. We then propose a categorization of deep reinforcement learning methodologies and discuss their advantages and limitations. In particular, we divide deep reinforcement learning into seven main categories according to their applications in computer vision, i.e. (i) landmark localization (ii) object detection; (iii) object tracking; (iv) registration on both 2D image and 3D image volumetric data (v) image segmentation; (vi) videos analysis; and (vii) other applications. Each of these categories is further analyzed with reinforcement learning techniques, network design, and performance. Moreover, we provide a comprehensive analysis of the existing publicly available datasets and examine source code availability. Finally, we present some open issues and discuss future research directions on deep reinforcement learning in computer vision.

The integration of learning and reasoning is one of the key challenges in artificial intelligence and machine learning today, and various communities have been addressing it. That is especially true for the field of neural-symbolic computation (NeSy) [10, 21], where the goal is to integrate symbolic reasoning and neural networks. NeSy already has a long tradition, and it has recently attracted a lot of attention from various communities (cf. the keynotes of Y. Bengio and H. Kautz on this topic at AAAI 2020, the AI Debate [9] between Y. Bengio and G. Marcus). Another domain that has a rich tradition in integrating learning and reasoning is that of statistical relational learning and artificial intelligence (StarAI) [39, 85]. But rather than focusing on integrating logic and neural networks, it is centred around the question of integrating logic with probabilistic reasoning, more specifically probabilistic graphical models. Despite the common interest in combining symbolic reasoning with a basic paradigm for learning, i.e., probabilistic graphical models or neural networks, it is surprising that there are not more interactions between these two fields.

For more than 30 years, Geoffrey Hinton hovered at the edges of artificial intelligence research, an outsider clinging to a simple proposition: that computers could think like humans do--using intuition rather than rules. The idea had taken root in Hinton as a teenager when a friend described how a hologram works: innumerable beams of light bouncing off an object are recorded, and then those many representations are scattered over a huge database. Hinton, who comes from a somewhat eccentric, generations-deep family of overachieving scientists, immediately understood that the human brain worked like that, too--information in our brains is spread across a vast network of cells, linked by an endless map of neurons, firing and connecting and transmitting along a billion paths. He wondered: could a computer behave the same way? The answer, according to the academic mainstream, was a deafening no. Computers learned best by rules and logic, they said. And besides, Hinton's notion, called neural networks--which later became the groundwork for "deep learning" or "machine learning"--had already been disproven. In the late '50s, a Cornell scientist named Frank Rosenblatt had proposed the world's first neural network machine. It was called the Perceptron, and it had a simple objective--to recognize images. The goal was to show it a picture of an apple, and it would, at least in theory, spit out "apple." The Perceptron ran on an IBM mainframe, and it was ugly.

As power grids fill up with renewable energy, electric vehicle charging stations and customer-owned generation, they will become too complex and fast-moving for their human operators to keep up with, a group of international researchers warns. The humans will need help from smart machines -- high-performance computers running decisionmaking software systems built with artificial intelligence -- according to researchers at France's grid operator RTE, the U.S. Electric Power Research Institute (EPRI) and other partners. With the proliferation of low-carbon options, "the grid becomes exponentially more challenging to operate," said Jeremy Renshaw, EPRI's AI director. "Grid operators are already stretched to the limit. Getting AI resources to help is going to be critical."

Multiclass multilabel classification refers to the task of attributing multiple labels to examples via predictions. Current models formulate a reduction of that multilabel setting into either multiple binary classifications or multiclass classification, allowing for the use of existing loss functions (sigmoid, cross-entropy, logistic, etc.). Empirically, these methods have been reported to achieve good performance on different metrics (F1 score, Recall, Precision, etc.). Theoretically though, the multilabel classification reductions does not accommodate for the prediction of varying numbers of labels per example and the underlying losses are distant estimates of the performance metrics. We propose a loss function, sigmoidF1. It is an approximation of the F1 score that (I) is smooth and tractable for stochastic gradient descent, (II) naturally approximates a multilabel metric, (III) estimates label propensities and label counts. More generally, we show that any confusion matrix metric can be formulated with a smooth surrogate. We evaluate the proposed loss function on different text and image datasets, and with a variety of metrics, to account for the complexity of multilabel classification evaluation. In our experiments, we embed the sigmoidF1 loss in a classification head that is attached to state-of-the-art efficient pretrained neural networks MobileNetV2 and DistilBERT. Our experiments show that sigmoidF1 outperforms other loss functions on four datasets and several metrics. These results show the effectiveness of using inference-time metrics as loss function at training time in general and their potential on non-trivial classification problems like multilabel classification.

There are some recent approaches and results about the use of answer-set programming for specifying counterfactual interventions on entities under classification, and reasoning about them. These approaches are flexible and modular in that they allow the seamless addition of domain knowledge. Reasoning is enabled by query answering from the answer-set program. The programs can be used to specify and compute responsibility-based numerical scores as attributive explanations for classification results.

Automatic Speech Recognition (ASR) is an active field of research due to its huge number of applications and the proliferation of interfaces or computing devices that can support speech processing. But the bulk of applications is based on well-resourced languages that overshadow under-resourced ones. Yet ASR represents an undeniable mean to promote such languages, especially when design human-to-human or human-to-machine systems involving illiterate people. An approach to design an ASR system targeting under-resourced languages is to start with a limited vocabulary. ASR using a limited vocabulary is a subset of the speech recognition problem that focuses on the recognition of a small number of words or sentences. This paper aims to provide a comprehensive view of mechanisms behind ASR systems as well as techniques, tools, projects, recent contributions, and possibly future directions in ASR using a limited vocabulary. This work consequently provides a way to go when designing ASR system using limited vocabulary. Although an emphasis is put on limited vocabulary, most of the tools and techniques reported in this survey applied to ASR systems in general.

We present a universally consistent learning rule whose expected error is monotone non-increasing with the sample size under every data distribution. The question of existence of such rules was brought up in 1996 by Devroye, Gy\"orfi and Lugosi (who called them "smart"). Our rule is fully deterministic, a data-dependent partitioning rule constructed in an arbitrary domain (a standard Borel space) using a cyclic order. The central idea is to only partition at each step those cyclic intervals that exhibit a sufficient empirical diversity of labels, thus avoiding a region where the error function is convex.

The transition from conventional methods of energy production to renewable energy production necessitates better prediction models of the upcoming supply of renewable energy. In wind power production, error in forecasting production is impossible to negate owing to the intermittence of wind. For successful power grid integration, it is crucial to understand the uncertainties that arise in predicting wind power production and use this information to build an accurate and reliable forecast. This can be achieved by observing the fluctuations in wind power production with changes in different parameters such as wind speed, temperature, and wind direction, and deriving functional dependencies for the same. Using optimized machine learning algorithms, it is possible to find obscured patterns in the observations and obtain meaningful data, which can then be used to accurately predict wind power requirements . Utilizing the required data provided by the Gamesa's wind farm at Bableshwar, the paper explores the use of both parametric and the non-parametric models for calculating wind power prediction using power curves. The obtained results are subject to comparison to better understand the accuracy of the utilized models and to determine the most suitable model for predicting wind power production based on the given data set.