Forward-looking statements are presented for the sole purpose of assisting investors and others in understanding Alithya's objectives, strategies and business outlook and may not be appropriate for other purposes. Although management believes the expectations reflected in Alithya's forward-looking statements were reasonable as at the date they were made, forward-looking statements are based on the opinions, assumptions and estimates of management and, as such, are subject to a variety of risks and uncertainties and other factors, many of which are beyond Alithya's control, and which could cause actual events or results to differ materially from those expressed or implied in such statements. Such risks and uncertainties include but are not limited to those discussed in Alithya's annual and interim Management's Discussion and Analysis and other materials made public, including documents filed with Canadian and U.S. securities regulatory authorities from time to time and which are available on SEDAR at www.sedar.com Additional risks and uncertainties not currently known to Alithya or that Alithya currently deems to be immaterial could also have a material adverse effect on its financial position, financial performance, cash flows, business or reputation.

To appear at NeurIPS 2019. Modular and programmable architecture search framework that allows you to implement your own search spaces and search algorithms through a consistent API. Reading the Twitter thread will give you a pretty good idea of the main ideas. Looking to get a few initial users and feedback.

Trint, the London-based transcription startup founded by Emmy-winning journalist Jeff Kofman, has raised $4.5 million in Series A funding. The round includes follow-on investment from Horizons Lab, the Hong Kong-based seed fund operated by the managers of Horizons Ventures, with participation from TechNexus, and The Associated Press. It brings total funding for Trint to $7.8 million since the company's founding in December 2014. Original backers include Google Digital News Innovation Fund and the Knight Enterprise Fund. Counting some of the world's largest media organizations as customers -- including The Associated Press, Vice News, The Washington Post and Der Spiegel -- Trint uses machine learning and speech-to-text technology to automate transcribing, which is a significant pain-point for journalists and other content producers, such as video makers.

MINE: Mutual Information Neural Estimation: The idea is quite simple - take a function that serves as a lower (or upper) bound for a quantity you want to estimate and use gradient ascent (or descent) to maximize (or minimize) the function. In this case, the function is a lower bound on KL divergence: Donsker-Varadhan representation. I have been thinking that this could become a more generalized approach to estimate values that are hard to evaluate from data. Basically take any inequality that involves the term we want to estimate, parameterize it and apply gradient descent/ascent.

As if the internet isn't already a complicated cesspool full of trolls, AI engineers have gone one step further to build a machine learning model that can generate fake comments for news articles. The eyebrow-raising creation, known as DeepCom, was developed by a group of engineers at Beihang University and Microsoft, China. "Automatic news comment generation is beneficial for real applications but has not attracted enough attention from the research community," they said in a paper released on arXiv. Allowing readers to post comments under articles keeps them engaged, they argued. Open dialogue allows people to discuss their opinions and share new information.

In cellular systems, the user equipment (UE) can request a change in the frequency band when its rate drops below a threshold on the current band. The UE is then instructed by the base station (BS) to measure the quality of candidate bands, which requires a measurement gap in the data transmission, thus lowering the data rate. We propose a band switching approach based on machine learning that does not require any measurement gap. Our proposed classifier-based band switching policy instead exploits spatial and spectral correlation between radio frequency signals in different bands based on knowledge of the UE location. We focus on switching between a lower (e.g. 3.5 GHz) band and a millimeter wave band (e.g. 28 GHz), and design and evaluate two classification models that are trained on a ray-tracing dataset. A key insight is that measurement gaps are overkill, in that only the relative order of the bands is necessary for band selection, rather than a full channel estimate. Our proposed machine learning-based policies achieve roughly 30% improvement in mean effective rates over those of the industry standard policy, while achieving misclassification errors well below 0.5%.