We consider the problem of deep neural net compression by quantization: given a large, reference net, we want to quantize its real-valued weights using a codebook with $K$ entries so that the training loss of the quantized net is minimal. The codebook can be optimally learned jointly with the net, or fixed, as for binarization or ternarization approaches. Previous work has quantized the weights of the reference net, or incorporated rounding operations in the backpropagation algorithm, but this has no guarantee of converging to a loss-optimal, quantized net. We describe a new approach based on the recently proposed framework of model compression as constrained optimization \citep{Carreir17a}. This results in a simple iterative "learning-compression" algorithm, which alternates a step that learns a net of continuous weights with a step that quantizes (or binarizes/ternarizes) the weights, and is guaranteed to converge to local optimum of the loss for quantized nets. We develop algorithms for an adaptive codebook or a (partially) fixed codebook. The latter includes binarization, ternarization, powers-of-two and other important particular cases. We show experimentally that we can achieve much higher compression rates than previous quantization work (even using just 1 bit per weight) with negligible loss degradation.

The aim of semi-supervised learning is to improve supervised learners by exploiting potentially large amounts of, typically easier to obtain, unlabeled data [1]. Up to now, however, semi-supervised learners have reported mixed results when it comes to such improvements: it is not always the case that semi-supervision results in lower expected error rates. On the contrary, severely deteriorated performances have been observed in empirical studies and theory shows that improvement guarantees can often only be provided under rather stringent conditions [2-5]. Now, the principal suggestion put forward in this chapter is that, when dealing with semi-supervised learning, one may not only want to study the (expected) error rates these classifiers produce, but also to measure the classifiers' performances by means of the intrinsic loss they may be optimizing in the first place. That is, for classification routines that optimize a so-called surrogate loss at training time--which is what many machine learning and Bayesian decision theoretic approaches do [6, 7], we propose to also investigate how this loss behaves on the test set as this can provide us with an alternative view on the classifier's behavior that a mere error rate cannot capture. In fact, though the main example is concerned with semi-supervision, we would like to argue that in other learning scenarios, similar considerations might be beneficial. For instance in active learning [8], where rather than sampling randomly from ones input data to provide these instances with labels, one aims to do the sampling in a systematic way, trying to keep labeling cost as low as one can or, similarly, to learn from as few labeled examples as possible. Also here it may (or, we believe, it should) be of interest to not only compare the error rates that different approaches (e.g.

We study a two-level multiview learning with more than two views under the PAC-Bayesian framework. This approach, sometimes referred as late fusion, consists in learning sequentially multiple view-specific classifiers at the first level, and then combining these view-specific classifiers at the second level. Our main theoretical result is a generalization bound on the risk of the majority vote which exhibits a term of diversity in the predictions of the view-specific classifiers. From this result it comes out that controlling the trade-off between diversity and accuracy is a key element for multiview learning, which complements other results in multiview learning. Finally, we experiment our principle on multiview datasets extracted from the Reuters RCV1/RCV2 collection.

For sophisticated reinforcement learning (RL) systems to interact usefully with real-world environments, we need to communicate complex goals to these systems. In this work, we explore goals defined in terms of (non-expert) human preferences between pairs of trajectory segments. We show that this approach can effectively solve complex RL tasks without access to the reward function, including Atari games and simulated robot locomotion, while providing feedback on less than one percent of our agent's interactions with the environment. This reduces the cost of human oversight far enough that it can be practically applied to state-of-the-art RL systems. To demonstrate the flexibility of our approach, we show that we can successfully train complex novel behaviors with about an hour of human time. These behaviors and environments are considerably more complex than any that have been previously learned from human feedback.

Prediction in a small-sized sample with a large number of covariates, the "small n, large p" problem, is challenging. This setting is encountered in multiple applications, such as precision medicine, where obtaining additional samples can be extremely costly or even impossible, and extensive research effort has recently been dedicated to finding principled solutions for accurate prediction. However, a valuable source of additional information, domain experts, has not yet been efficiently exploited. We formulate knowledge elicitation generally as a probabilistic inference process, where expert knowledge is sequentially queried to improve predictions. In the specific case of sparse linear regression, where we assume the expert has knowledge about the values of the regression coefficients or about the relevance of the features, we propose an algorithm and computational approximation for fast and efficient interaction, which sequentially identifies the most informative features on which to query expert knowledge. Evaluations of our method in experiments with simulated and real users show improved prediction accuracy already with a small effort from the expert.

Answer Set Programming (ASP) is a well-established declarative paradigm. One of the successes of ASP is the availability of efficient systems. State-of-the-art systems are based on the ground+solve approach. In some applications this approach is infeasible because the grounding of one or few constraints is expensive. In this paper, we systematically compare alternative strategies to avoid the instantiation of problematic constraints, that are based on custom extensions of the solver. Results on real and synthetic benchmarks highlight some strengths and weaknesses of the different strategies. (Under consideration for acceptance in TPLP, ICLP 2017 Special Issue.)

With the advent of artificial general intelligence and self-designed intelligent programs, new and more intelligent AI will appear, rapidly creating ever smarter machines that will, eventually, surpass us. When we reach this so-called AI singularity, our minds and bodies will be obsolete. Humans may merge with machines and continue to evolve as cyborgs. Is this really what we have to look forward to? AI, a scientific discipline rooted in computer science, mathematics, psychology, and neuroscience, aims to create machines that mimic human cognitive functions such as learning and problem-solving.

The 2018 Audi A8, just unveiled in Barcelona, counts as the world's first production car to offer Level 3 autonomy. Level 3 means the driver needn't supervise things at all, so long as the car stays within guidelines. Here that involves driving no faster than 60 kilometers per hour (37 mph), which is why Audi calls the feature AI Traffic Jam Pilot. Go ahead, Audi's saying, read your newspaper or just zone out while traffic creeps along. Take a look at the company's promotional video.

After helping to launch the Partnership on AI with Google, Facebook and others; and doubling down on AI research, today Microsoft unveiled a new initiative that points to how it plans to target specific verticals in what can potentially be a very nebulous field -- while also raising the public image of AI as some grow concerned about the implications of its encroaching influence. Today, the company announced AI for Earth, a new program that will be dedicated to AI-based projects in the areas of agriculture, water, biodiversity and climate change, where Microsoft proposes to donate up to $2 million in Microsoft tools, services and training per project to help them get a leg up. It will be led by Microsoft Chief Environmental Scientist Lucas Joppa. AI for Earth will not include money grants as such -- although as recent investments from Microsoft Ventures show (and also its acquisitions), the company is also willing to put money where its mouth is, and to take equity in the most interesting projects, too. This program could help create a funnel to identify and follow some of the more interesting projects (and get them using Microsoft services to boot).

Handwriting will never be the same again. A new glove developed at the University of California, San Diego, can convert the 26 letters of American Sign Language (ASL) into text on a smartphone or computer screen. Because it's cheaper and more portable than other automatic sign language translators on the market, it could be a game changer. People in the deaf community will be able to communicate effortlessly with those who don't understand their language. ASL is a language all of its own, but few people outside the deaf community speak it.