In this paper we address the challenge of land cover classification for satellite images via Deep Learning (DL). Land Cover aims to detect the physical characteristics of the territory and estimate the percentage of land occupied by a certain category of entities: vegetation, residential buildings, industrial areas, forest areas, rivers, lakes, etc. DL is a new paradigm for Big Data analytics and in particular for Computer Vision. The application of DL in images classification for land cover purposes has a great potential owing to the high degree of automation and computing performance. In particular, the invention of Convolution Neural Networks (CNNs) was a fundament for the advancements in this field. In [1], the Satellite Task Team of the UN Global Working Group describes the results achieved so far with respect to the use of earth observation for Official Statistics. However, in that study, CNNs have not yet been explored for automatic classification of imagery. This work investigates the usage of CNNs for the estimation of land cover indicators, providing evidence of the first promising results. In particular, the paper proposes a customized model, called Satellite-Net, able to reach an accuracy level up to 98% on test sets.

We present multi-modal adversarial autoencoders for recommendation and evaluate them on two different tasks: citation recommendation and subject label recommendation. We analyze the effects of adversarial regularization, sparsity, and different input modalities. By conducting 408 experiments, we show that adversarial regularization consistently improves the performance of autoencoders for recommendation. We demonstrate, however, that the two tasks differ in the semantics of item co-occurrence in the sense that item co-occurrence resembles relatedness in case of citations, yet implies diversity in case of subject labels. Our results reveal that supplying the partial item set as input is only helpful, when item co-occurrence resembles relatedness. When facing a new recommendation task it is therefore crucial to consider the semantics of item co-occurrence for the choice of an appropriate model.

The unreasonable effectiveness of data is possibly the greatest surprise coming out of the last twenty years of Artificial Intelligence (AI): pretty simple algorithms and tons of data seem to almost invariably beat complex solutions with small-to - none training set. In the seminal words of (Halevy, Norvig, and Pereira, 2009): "now go out and gather some data, and see what it can do". The perfect storm has been set in motion by the convergence of the big data hype (Hagstroem et al 2017), the general availability of specialized hardware and scalable infrastructure, and some "computational tricks" (e.g. Hochreiter S., Schmidhuber S., 1997, Hinton et al, 2013): all together, they unlocked the Deep Learning (DL) Revolution and created a tremendous amount of business value (Chui et al 2018). The A.I. wave is so disruptive that a great deal of commentators, practitioners (Radford et al 2019) and entrepreneurs (Musk 2017) inevitably started to wonder what is the place of humans in this new world: is A.I. going to replace humanity (in the world of Silicon Valley, Joy in 2001 was already stating that "the future doesn't need us")? In this position paper, we shall argue for two surprising perspectives: 1) the future of A.I. is about less data, not more; 2) human-machine collaboration is, at least for the foreseeable future, the only way to outpace humans and outsmart machines effectively. The paper is organized as follows: Section 2 contains a review of the current state of the A.I. landscape, with particular attention to the origins of the DL Revolution; the section casts some doubts on the general applicability of DL to language problems, drawing from theoretical considerations from academia and industry use cases in the space of Tooso. Section 3 details a real use-case from the industry that is challenging for the DL paradigm, and outlines a different framework to tackle the problem; finally, Section 4 concludes with remarks and roadmap for a new type of A.I., what we call "A.I. with humans and for humans." 2

Neural generation methods for task-oriented dialogue typically generate from a meaning representation that is populated using a database of domain information, such as a table of data describing a restaurant. While earlier work focused solely on the semantic fidelity of outputs, recent work has started to explore methods for controlling the style of the generated text while simultaneously achieving semantic accuracy. Here we experiment with two stylistic benchmark tasks, generating language that exhibits variation in personality, and generating discourse contrast. We report a huge performance improvement in both stylistic control and semantic accuracy over the state of the art on both of these benchmarks. We test several different models and show that putting stylistic conditioning in the decoder and eliminating the semantic re-ranker used in earlier models results in more than 15 points higher BLEU for Personality, with a reduction of semantic error to near zero. We also report an improvement from .75 to .81 in controlling contrast and a reduction in semantic error from 16% to 2%.

Modern deep learning methods promised to relegate this practice to history, but have not eliminated the interest in sparse modeling for NLP . Along with concerns about computational resources (Chen et al., 2016; Narang et al., 2017b) and interpretability (Murphy et al., 2012; Subramanian et al., 2018), human intuitions continue to motivate sparse representations of language. For example, some work applies assumptions of sparsity to model latent hard categories such as syntactic dependencies (Padó and Lapata, 2007) or phonemes (Cotterell and Eisner, 2018). Niculae and Blondel (2017) found that a sparse attention mechanism outperformed dense methods on some NLP tasks; Narang et al. (2017a) found sparsified versions of LMs that outperform dense originals. Attempts to engineer sparsity rest on an unstated assumption that it doesn't arise naturally when neural models are learned.

Today's interactive devices such as smart-phone assistants and smart speakers often deal with short-duration speech segments. As a result, speaker recognition systems integrated into such devices will be much better suited with models capable of performing the recognition task with short-duration utterances. In this paper, a new deep neural network, UtterIdNet, capable of performing speaker recognition with short speech segments is proposed. Our proposed model utilizes a novel architecture that makes it suitable for short-segment speaker recognition through an efficiently increased use of information in short speech segments. UtterIdNet has been trained and tested on the V oxCeleb datasets, the latest benchmarks in speaker recognition. Evaluations for different segment durations show consistent and stable performance for short segments, with significant improvement over the previous models for segments of 2 seconds, 1 second, and especially sub-second durations (250 ms and 500 ms).

Fact verification (FV) is a challenging task which requires to retrieve relevant evidence from plain text and use the evidence to verify given claims. Many claims require to simultaneously integrate and reason over several pieces of evidence for verification. However, previous work employs simple models to extract information from evidence without letting evidence communicate with each other, e.g., merely concatenate the evidence for processing. Therefore, these methods are unable to grasp sufficient relational and logical information among the evidence. To alleviate this issue, we propose a graph-based evidence aggregating and reasoning (GEAR) framework which enables information to transfer on a fully-connected evidence graph and then utilizes different aggregators to collect multi-evidence information. We further employ BERT, an effective pre-trained language representation model, to improve the performance. Experimental results on a large-scale benchmark dataset FEVER have demonstrated that GEAR could leverage multi-evidence information for FV and thus achieves the promising result with a test FEVER score of 67.10%. Our code is available at https://github.com/thunlp/GEAR.

The field of artificial intelligence (AI) represents an enormous endeavour of humankind that is currently transforming our societies down to their very foundations. Its task, building truly intelligent systems, is underpinned by a vast array of subfields ranging from the development of new electronic components to mathematical formulations of highly abstract and complex reasoning. This breadth of subfields renders it often difficult to understand how they all fit together into a bigger picture and hides the multi-faceted, multi-layered conceptual structure that in a sense can be said to be what AI truly is. In this perspective we propose a system of five levels/layers of abstraction that underpin many AI implementations. We further posit that each layer is subject to a complexity-performance trade-off whilst different layers are interlocked with one another in a control-complexity trade-off. This overview provides a conceptual map that can help to identify how and where innovation should be targeted in order to achieve different levels of functionality, assure them for safety, optimise performance under various operating constraints and map the opportunity space for social and economic exploitation.

Recent studies on automatic neural architectures search have demonstrated significant performance, competitive to or even better than hand-crafted neural architectures. However, most of the existing network architecture tend to use residual, parallel structures and concatenation block between shallow and deep features to construct a large network. This requires large amounts of memory for storing both weights and feature maps. This is challenging for mobile and embedded devices since they may not have enough memory to perform inference with the designed large network model. To close this gap, we propose MemNet, an augment-trim learning-based neural network search framework that optimizes not only performance but also memory requirement. Specifically, it employs memory consumption based ranking score which forces an upper bound on memory consumption for navigating the search process. Experiment results show that, as compared to the state-of-the-art efficient designing methods, MemNet can find an architecture which can achieve competitive accuracy and save an average of 24.17% on the total memory needed.

The success of deep learning in medical imaging is mostly achieved at the cost of a large labeled data set. Semi-supervised learning (SSL) provides a promising solution by leveraging the structure of unlabeled data to improve learning from a small set of labeled data. Self-ensembling is a simple approach used in SSL to encourage consensus among ensemble predictions of unknown labels, improving generalization of the model by making it more insensitive to the latent space. Currently, such an ensemble is obtained by randomization such as dropout regularization and random data augmentation. In this work, we hypothesize -- from the generalization perspective -- that self-ensembling can be improved by exploiting the stochasticity of a disentangled latent space. To this end, we present a stacked SSL model that utilizes unsupervised disentangled representation learning as the stochastic embedding for self-ensembling. We evaluate the presented model for multi-label classification using chest X-ray images, demonstrating its improved performance over related SSL models as well as the interpretability of its disentangled representations.