Shulou(Shulou.com)06/02 Report--

Produced by big data Digest

The high-profile NeurlPS 2019 kicked off early Sunday morning in Vancouver. Today, the organizing committee announced five categories of awards, including the Outstanding thesis Award and the Outstanding New Direction Paper Award, through a blog.

NeurIPS, the full name of Neural Information processing Systems Conference (Conference and Workshop on Neural Information Processing Systems), has a history of 32 years since it was born in 1987. It has been highly concerned by academia and industry. NeurIPS is a Class A conference in the field of artificial intelligence, and it is also one of the most famous annual conferences in the field of artificial intelligence.

According to official sources, the number of papers contributed to this year's NeurIPS conference reached a new high, receiving a total of 6743 contributions, resulting in 1428 papers and 36 oral, with an acceptance rate of 21.2%.

The number of participants this year has also reached a new high, with the total number of participants exceeding 13000, an increase of nearly half compared with the 9000 participants last year. Due to the excessive number of applicants. Organizers of this year's conference even changed the rules of ticket grabbing, changing the "first-come-first-served" principle to the "lucky draw" principle. As NeurIPS said in the announcement, NeurIPS 2019, which grabs tickets with the new rules, can avoid time zone problems to some extent.

The registration queue for this year's opening ceremony, picture: HooraRM @ NeurIPS2019

And such a high contribution rate and attendance rate can also reflect from the side that the popularity of artificial intelligence research is only increasing.

This year's participation commemorates the mug.

With such a large amount of contributions, the organizing committee also faces a lot of pressure in the process of re-examination and selection. In the official blog post announcing the award, NeurlPS 2019 also states the relevant criteria and selection process in the award review process, from which we can see which studies are more recognized or discouraged by NeurIPS 2019.

The evaluation criteria unanimously endorsed by the review committee include encouraging the following nine research characteristics:

Long-term potential focuses on the main work, not on the corners and corners. People may still be concerned about this issue in the coming decades.

Insight-provide new (preferably in-depth) understanding; not just to show a few percentage points of improvement in the model

Creativity-looking at problems in a creative new way, giving a result that really surprises the reader

Revolutionary thinking will fundamentally change people's way of thinking in the future.

Unassailable rigor and consideration.

Elegant feeling beautiful, clean, smooth, elegant

Reality does not overemphasize its importance.

Scientific-- it can be falsified in practice.

Repeatable examples the results are reproducible; the code is available and can be used on a variety of machines; the dataset is available; and the details are complete.

The organizing committee also announced some of the three characteristics of the paper that they do not encourage:

Inefficient workers stay away from jobs that stand out simply because of a waste of resources (mainly because they waste a lot of resources and get a higher ranking)

Follow the trend-use one method because an idea is fashionable, but you can get it in a different and more effective way in other ways.

There is unnecessary complexity in a paper that is too complex.

In addition, the Organizing Committee has added a new Outstanding New Direction Paper Award this year to highlight its excellent work in setting a new approach for future research.

Let's take a look at the winning papers of the five major awards.

Outstanding thesis Award (Outstanding Paper Award)

Award-winning papers:

Distribution-Independent PAC Learning of Halfspaces with Massart Noise

Links to papers:

Https://papers.nips.cc/paper/8722-distribution-independent-pac-learning-of-halfspaces-with-massart-noise.pdf

Authors: Ilias Diakonikolas, Themis Gouleakis, Christos Tzamos

Institution: University of Wisconsin-Madison, Mapu Institute

Specifically, the researchers gave a set of labeled samples (x, y) to sample R ^ d + 1-dimensional distribution D, so that the edge distribution on the untagged point x is arbitrary, and the label y is generated by an unknown half-space, which is destroyed by Massart noise, where the noise rate η

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.