The O’Reilly Artificial Intelligence Conference 2019, Beijing, China

About this course

so we wanted to take a few minutes to highlight the results some of some recent surveys and along the way describe some trends in AI and highlight

some of the upcoming talks at the conference so first as many of you are aware a AI and machine learning are extremely active research areas and for the most part that research is being

done out in the open the website archive.org a very popular online website for preprints had an average daily upload of a hundred machine learning papers in 2018 and all that

research is being translated to patent filings I was fortunate enough to take part in a recent study by the World Intellectual Patent Office and I say you can see the rate of patent filings is

also growing very fast and this reinforces the notion that we have entered the implementation phase for AI technologies in fact there was a six-fold increase in annual patent

filings over the 12-year period spanning from 20 from 2006 to 2017 and in fact 53% of all the patents in AI were filed after 2013 drilling down into specific areas computer vision as many of you are

aware there are many startups and many commercial use cases to have computer vision and we will have a lot of sessions about computer vision at this conference now it's been a extremely

productive year for researchers in natural language every few months it seems like there seem to be new deep learning models that established records in many different natural language tasks

and benchmarks much of this research was done in the open accompanied by open source code of pre trained models which is great for the entire community and we are already

seeing companies begin to tune these models for their specific domains and applications there are many sessions on text speech and natural language at this conference including a talk on transfer

learning today by David Lowe of panned AI we continue to see treatments and tools for deep learning as well our surveys show that pencil and pad Taurus remain in the most popular

libraries now what framework are practitioners and companies using a recent study by UC Berkeley Ryze lab shed some light on this question for each monk they looked at the number of

questions posted on Stack Overflow tagged with each framework Stack Overflow if you recall is a question and answer site for programmers tensorflow is clearly the more popular framework

from January 2017 through the end of April of this year of 2019 in this afternoon quezon Wang of Google will give a talk outlining what's new for tensorflow 2.0 now what framework are

researchers using the same study by used to burglarize lab shed some light on this question so for each month they looked at the number of papers now post on our archive that mention each

framework and from January 2017 through December 2018 tensorflow was the most cited work among researchers but things are beginning to shift by April 2019 a PI torched surpassed tensor flow to the

extent that researchers are early indicators look for PI to also gain users among practitioners and companies joseph is AK of Facebook will be joining us and he will give an update on the

current state of Pi torch tomorrow morning so in our recent survey that drew over 1300 respondents about 22% signaled that they are beginning to use reinforcement learning so there are many

reasons for this but one of the main reasons it's that we just have more tools open source for Prai Terry and cloud services for doing RL so we have a couple of keynote speakers over the next

few days and we'll talk about reinforcement learning including professor Yan steiker from UC Berkeley Rice lab and Professor Tim Kroshka of MIT last year we began

tracking many startups building specialized hardware for deep learning and AI both for training and inference and both for use in edge devices as well as data centers towards the latter part

of this year we expect some of these company to began releasing hardware that would greatly accelerate both training and inference while being much more energy

efficient at this conference we have a few talks on next-generation hardware including keynotes by a Tom Medina Habana and micro James a cerebral now this is all in the beginning to

translate into actual investments now in our past recent survey we found out over 60% of companies were planning to invest at least some of their IT budget into AI but the level of investment depends on

how much experience a company already has with AI technology as you can see from this slide those two mature practices plan to invest a small sizable portion of their IT budget into AI and

as a result there's a strong likelihood that the gap between leaders and laggards were widen in the space so what is holding back adoption of AI so according to our survey the answer

depends also on the level of maturity those who are getting started still struggle with finding use cases or explaining the importance of AI and machine learning to their fellow

employees on the other hand those with mature practices struggle with things like lack of data and lack of skilled people as the use of AI technologies grows within companies we will need

better tools for machine learning development and tools that can automate many stages of a machine learning pipeline at this conference we have many talks on automation and data platforms

including a talk on Auto ml by cui sway of Microsoft this afternoon and a talk from racket and tomorrow on best practices for building a data science team now with this shift towards the

implementation of machine learning it's natural to expect that tools targeted helping companies with with ml practices will start to improve as well as you can see we are beginning to track the many

companies and open source projects for model development governance and operations and these are all active areas for researchers and entrepreneurs so expect rapid progress in the next 12

months now companies are realizing that machine learn a I involve much more than optimizing individual business or statistical metrics and we have many speakers or

over the next two days who will explain various techniques that are being used to develop responsible AI in one recent survey we found many companies are beginning to check for things like

explained ability fairness and bias security and privacy safety and reliability in fact many of these topics are very active areas of research and much of that research is being

translated into new tools and best practices we have a few talks on these topics including a session this afternoon on federated learning by Alex Ingram in of Google so check that out we

are also seeing companies begin to retrain their workforce on topics pertaining to AI and machine learning so here are some high-level results from a recent analysis we did of content usage

and O'Reilly learning platform so for those of you are not familiar subscribers to O'Reilly learning have access to books live online training and video so let's examine the largest

topics on all Riley learning in terms of content usage from January to May 2019 in this chart the size of a square is a proxy for the popularity of a topic the color of square reflects the rate of

growth where green means the topic is growing and red means the topic smaller in 2019 compared to what it was in 2018 and as you can see there are many topics related to AI machine learning and data

among the most popular topics now among the trending topics which we also see quite a few that are related to AI machine learning and data so all of these are early science that companies

and employees are beginning to update their skills for the AI and machine learning era to summarize we are in the implementation phase for AI machine learning tools are improving use cases

and case studies are being shared more widely and companies and employees are beginning to update their skills as we noted earlier once ina companies are taking AI

seriously is to look at the growth and patents computer vision which is the topic of several keynotes and such as here over the next two days is by far the most popular application area for

current AI technologies many companies have long collected text as well and many have experienced we have using MLP NLP and text mining tools recent progress and deepening for language

models will also surely result in many new and interesting applications within that space and don't forget that we are still in the early days for speech applications but even at this stage we

are already seeing many interesting consumer and enterprise applications emerge so we are still in the early stages of machine learning and AI within companies tools to trap and manage

experiments and ant and machine learning pipelines are beginning to appear we look forward to tools for building the AI applications of tomorrow and I encourage you to walk around the expo

hall as many of our sponsors might already have what you need and in fact we have an academic project here at the conference from MIT called Northstar it's located near the O'Reilly booth so

I encourage you to check out this MIT project so we have another outstanding program for you over the next few days there will be many sessions on topics that are critical to sustaining the use

of AI within your organization's as much as we love the programís conference organizers we encourage you to network with other people at the conference so so with speakers with the companies in

the expo hall as well as with your fellow attendees and just to highlight the importance of networking we've arranged so that many of our keynote speakers will make themselves available

at lunch for informal discussions so for the next few days this is your AI community you