A Year In The Life Of RE•WORK: Summit Highlights

Original

Having hosted our final summit of 2018 at the end of November, we’ve had some time to take a look back at the fantastic year RE•WORK has had. 2018 marks the most RE•WORK events ever hosted with us taking 12 topics to 8 cities in 4 countries. Kicking off in the US, heading to Europe, Asia, Canada and back to the states has been a whirlwind to say the least.

This year we hosted new topics such as AI in Industrial Automation, and AI for Government and were welcomed to new cities such as Houston and Toronto. Some of the world leaders in AI have honoured not only us, but the whole community by sharing their pioneering work at our summits including Geoffrey Hinton, University of Toronto; Ian Goodfellow, Google Brain; Daphne Koller, Calico;  Jeff Clune, Uber AI Labs; Raia Hadsell, DeepMind; Georgia Gkioxari, Facebook AI Research (FAIR) and many more.

We’re now looking forward to the world’s biggest Deep Learning Summit in San Francisco taking place this January 24 - 25, but before we wave goodbye to 2018, we wanted to share some of the learnings and highlights of the year:


Deep Learning Summit & AI Assistant Summit

  • Our goal is to make human-computer interactions more personal. Examples of these personalised generative models stretch to personalised voice assistants, avatars and virtual reality. Yaniv Taigman, FAIR

  • Controversially I think C++ coders are going to be replaced long before truck drivers. I think autonomous vehicles are at the far end of AI. If a car needs to know how to drive in India, it needs to understand so many things about the road, like How an elephant moves in traffic! Keith Adams, Slack

  • In my day to day life at Google I’m mainly used to working with people familiar with solving machine learning problems, but here I get to come and see 'what do people in the business world want machine learning researchers to solve for them?' Ian Goodfellow, Google Brain

  • Computers are getting better and better at clearly defined pattern recognition tasks, we can do problems we couldn’t before. However, if you take a programme that can recognise images and ask it to do something with natural language and it’s stumped! We’re nowhere near general AI. Daphne Koller, Calico

  • The idea of hindsight experience replay in reinforcement learning Is that it never wastes any experience. Ilya Sutskever, OpenAI

  • For us, as a healthcare company, we think a lot about patient data privacy. In a broader sense, we want to make sure that we’re ethical in the way we get our patients to be compliant. Cathy Pearl, Sense.ly

  • Diversity in the collection of datasets is important. Accents with voice recognition software for example, If you train the system on a wide set of accents it’s more likely to perform accurately for a wider audience. Abhishek Gupta, Microsoft

  • We have deployed hundreds of imaging satellites in space to capture nearly the entire Earth’s landmass every day. This unprecedented data feed enables countless applications to detect, quantify and understand how our world is changing. Jesus Martinez Manso. Planet


Deep Learning in Finance Summit, Deep Learning in Retail & Advertising Summit, AI Assistant Summit

  • We need to find a way to get people at Prudential to understand that there’s a better way to do things with AI rather than the standard ways we’ve always done things, and this is ultimately to get better customer experiences for our users. Michael Natusch, Prudential

  • The next tech revolution is happening now and it’s powered by AI. We’re shifting to a world where machines are adapted to humans. Yariv Adan, Google Assistant

  • I have seen the massive ability AI has to change the world for good, and we must continue to create AI in an open, transparent way. Kriti Sharma, Sage

  • When Skyscanner is trying to plot the best travel itinerary, it can’t get the complexity of all possible variants, so we identify the top 10 - to do this we train a classifier for a given query to predict whether it’s going to be in the top 10 best value option or not. Dima Karamshuk, Skyscanner

  • Everyone building an AI product has to be empathic because that will ultimately feed down into the UX. Speak aloud what you’re designing - if you wouldn’t say it, your VA shouldn’t either. Put yourself in your users shoes and that’s the first step towards empathic design. Marc Paulina, Google

  • We heavily use machine learning in product recommendation. Say you’re looking at a pair of jeans and we want to recommend you something similar so we do this based on user history, and each product can be abstracted using an embedding in word2vec. Alessandro Magnani, Walmart Labs

  • AI can automate and optimise pretty much every single process you can think of because unlike most of the processes used in finance it can adapt to local scale and larger scale. Javier Campos, Experian

  • How can we compute data without exposing it? Let’s explore emerging cryptographic methods like Secure Multiparty Computation. Jordan Brandt, Inpher


Deep Learning Summit

  • If you work in ML today, you almost certainly work in Python. But it doesn't mean it should end there. ... The web world has a lot to offer for ML... With Tensorflow.js, we are building a new ecosystem of Machine Learning in browser. Daniel Smilkov, Google Brain

  • Labeled data is new oil. The quality gap in AI products between small and medium companies and industry giants is due to difference in amount of labeled data. Sudeep Pillai, Toyota Research Institute

  • Netflix’s models understands differences in local taste, takes into account availability and context. We use deep models to come up with personalized recommendations. Anoop Deoras, Netflix

  • AI is more like a bionic arm than a truckload of robots taking over the world. It’s a helper, not a replacer. Tom Wilde, Indico

  • A lot of things that are happening on the cloud, we think should be happening directly on the device. This means true privacy- your data never has to leave your device in order to. Chris Lott, Qualcomm Technologies

  • We’re never going to create AI with digital. Why? Because our brains are not digital. Bill Aronson, Artificial Intelligence Research Group

Deep Learning in Healthcare Summit
  • There are so many time consuming and menial tasks in healthcare. At Philips we’re providing a way to bring information from different domains together and also fuse and correlate them to help interpret information cross domain. Amir Tahmasebi, Philips

  • Our team is interested in taking genomic data and using tensorflow to solve the problem. Cory McLean, Google Brain

  • Clinical notes is where the information we find is most important. Whilst there’s more ‘stuff’ there, it's more difficult to process - not everything is structured so it’s important that we use structured data and clinical notes. Ricardo Mitto, Icahn School of Medicine

  • I’ll show you today how we use clinical NLP to process and recognise all the entities in a sentence and identify their semantic types and translate them into useful information. Chen Lin, Boston Children's Hospital

  • We have a lot of data on a large population of very sick people, as they’re chronic issues there’s a lot of treatment that needs to be provided which is expensive, so we have to be resourceful. Tommy Blanchard, Fresenius Medical Care

  • Over 70% of clinical decisions are based on lab data. But this data wasn’t created for analytical purposes (e.g., no standard format). It’s a mess! Prognosis is addressing how to make the data usable at scale. Fernando Schwartz, Prognos AI


Deep Learning for Robotics Summit & AI in Industrial Automation Summit

  • Imitation learning is learning from demonstration - we want robots to learn without active demonstrations. The most appealing learning for us is self-supervised learning where you combine the rich features of supervised learning, but without the labels. Corey Lynch, Google Brain

  • Humans don’t just passively process data - we explore, and so we need to teach intelligent agents to develop an intelligence to understand the concept of learning interactively. Georgia Gkioxari, FAIR

  • One car is manufactured every 3 seconds at Renault, Nissan, Mitsubishi. Even for 10 million cars, we will only have 10 defective parts. Our target is clear - let’s try to design something that would use a limited data set and limited time. Lionel Cordesses, Renault

  • We want systems to work outside of the lab, in the messiness of the real world, and to go beyond and perform the tasks at a complexity that has not been shown today. Animesh Garg, Stanford University AI Lab

  • AI without data is like a bike without wheels - if you don’t have the data in place, the algorithms won’t get you anywhere. Applied AI gives you ML, Domain Knowledge and Data - if any one of those is missing, the whole operation falls over. Shameer Miraz, PepsiCo

  • One of the big challenges is that we are interested in detecting particular instances of objects, for which there is little data available. Jana Kosecka, Google

Women in AI Dinner
  • Accurate user issue selection is very important as it helps us determine the material we expose to the customer or agents to resolve the issue. If we can automatically identify the issue we can free our customers and agents from mapping customer's problems to our internal vocabulary of potential issues and get to supporting our customers more quickly. Negin Nejati, Airbnb

  • If there’s a blind user who can’t see an image, but you can connect it to language it can help break down the intersection of vision and language. Devi Parikh, Facebook (FAIR)

  • Uber is creating numerous jobs, allowing people more freedom to move to a new city and start afresh with a new job instantaneously. Rosanne Liu, Uber AI Labs


Machine Intelligence Summit & AI in Healthcare Summit

  • E-commerce is exciting but using AI to develop smart cities is important as it touches everyone in their daily lives. Alan Lu,eBay

  • Healthcare is the number one industry application of AI in terms of revenue. This is because you do not have to prove the value of the product to anyone. Increased automation to save time in the healthcare industry is of paramount importance. Efstratios Tsougenis, Insight Medical Technology

  • In an AI age, data is key. Partnerships are necessary to drive innovation and advancement in an environment where data is not readily available. With a look towards the future, I believe blockchain solutions will solve data sharing problems as it increases confidence in data governance. Blockchain, AI and healthcare should go hand in hand. Lawrence Wee, Allianz Asia Pacific

  • There is a need to consider both the cultural and technical when implementing AI successfully.  Michael Natusch, Prudential

  • In the healthcare industry in particular, we need to increase collaboration to bring together the best technology from all over the world. Raymond Tong, CUHK


Women in AI Dinner

  • We know the challenges of understanding a word we have never seen before. Attempts to capture the meaning of words in a computational sense is important. Aleksandra Piktus, Facebook

  • We are trying to design an incentive with your preferences in mind to warrant a change in your behaviour in order to end up with the preferred solution. The challenge, however, is “How can we try to get agents to make better decisions?”. Sofia Ceppi, PROWLER.io

  • What takes 15 years and $3bn to create? This isn’t the start of a bad joke, this is how long, and how much money it takes to bring a new drug to market. Noor Shaker, CTN


Women in AI Reception
  • What is magic? Magic at Spotify is making a recommendation system that makes it feel like we can read user’s minds. Catie Edwards, Spotify

  • Machine translation doesn’t fully cover text to speech due to things like dialect. It can misunderstand the semantic context. Julia Kroll, Amazon Alexa

  • Human created data is a treasure trove of opportunity. We can then try to build a model that mimics human behaviour. Lucy X Wang, BuzzFeed

AI in Finance Summit
  • There are nine killer applications of digital technology in general insurance, and there are several Deep Learning practices in insurance: here, there are three major applications. These are centred around speech, text and NLP. Yuanyuan Liu, AIG

  • Deep neural networks deliver exceptional performance on tasks including credit fraud detection, malware classification, but we must not forget that it is incredibly easy to fool DNN’s. Roxana Geambasu, Columbia University

  • Most chatbots today are not robust, and fail in most cases as they cannot deal with the infinite capacity of language. National Bank of Canada are using Rasa Stack, an open source model to build the chatbot to overcome this. Eric Charton, National Band of Canada

  • Bias Variance can help you save time and cost. Data is not cheap. You need to pay a lot of money to train your algorithms. Francis Z Lin, BNY Mellon

  • When considering auditing machine learning, it is important that we look at it holistically, collaborating with different teams over an organisation. One tool that Capital One has developed is an “unsupervised learning data exploration environment”, in the face of audit’s problem of having lots of data in different places. Andrew Clark, Capital One


Deep Learning Summit
  • There is an elephant in the room. The problem is that the situation (digital maze) is not real, therefore we have to look into the agent learning of navigation in a real-life situation. Raia Hadsell, DeepMind

  • The computer vision industry has improved incredibly over the last 6 years. Scene recognition can now be done with 75% accuracy, which 10 years ago was unheard of. Agata Lapedriza, Universitat Oberta de Catalunya & MIT

  • During a hurricane getting information to emergency responders is crucial, essentially this project created a fusion which gives an idea of what is happening inside a hurricane at ground level. Deep learning is crucial for so many things at FDL. James Parr, NASA Frontier Development Lab

  • When a search engine delivers a biased result, should it be altered in order to deliver a desired result for the public, or an accurate result that may not be desired? Cansu Canca, Ethics AI Lab

  • In order to make a visual prediction, the model must know where the objects are positioned in 3d space, must understand depth perception, and must understand relative positioning. All of this is done without engineers, and solely with data. Ali Eslami, DeepMind

  • When going into meetings I push the idea that without data you are just another person with an opinion. Angel Serrano, Santander

Deep Learning in Healthcare Summit
  • We’re researching into detection of cancers and predicting patient deterioration. Over 80% of all vision impairments could be avoided if detected earlier. This opened up a huge opportunity for Deep Learning to be applied. Trevor Back, DeepMind

  • What we are able to do is take tens of thousands of CT scans, take all of this data and run automatic segmentation tools which create mesh’s and then in turn training data. Peter Mountney, Siemens Health

  • One of the main challenges in medicine is that highly trained human experts disagree with data on data instances. In more than 21% of cases there is significant disagreement. We study the application of ML to predict which instances are likely to give rise to maximal expert disagreement. Maithra Raghu, Google Brain

  • We started a project working on lung cancer and we use a different approach because of the scale of sample sets we have. There is a shortage of research pathologists in the country and so we developed this web system by ourselves. Yinyin Yuan, The Institute of Cancer

  • Research Whilst AI can solve countless problems, we need to educate the clinician community on what AI can’t do, not what it can do. Mark Gooding, Mirada


Women in AI Dinner

  • There’s room for more women in AI, we’d all agree, and everyone in this room agrees. I’m here to learn from everyone in this room, and I hope we can all learn from each other. Alison Paprica, Vector Institute

  • Data is inherently biased - data is given by humans. AI algorithms are created by biased humans I’m an optimist so I’ll get to that later!- but we shouldn’t be biased. Jennifer Gibbs, TD Bank

  • I’ve been working in linguistics, even though my background is in computer science, language has always fascinated in language has a strong connection to human intelligence, and therefore Artificial Intelligence. Afsaneh Fazly, Samsung Toronto AI Research Center

  • WinterLight Labs are creating AI applications to employ ML models to detect these diseases by looking at tests. We want to create something more robust system. We believe it’s possible to create a system that’s robust enough to identify cognitive disease through NLP. Jekaterina Novikova, WinterLight Labs

Deep Learning Summit

  • We’ve come so far, but there are still challenges - for example, when we layer images systems get confused - something that wouldn’t confuse humans. These systems are supposed to be superhuman. David Cox, IBM

  • Put up your hand if you’re certain if your definition of what an interpretable model is, is the same as your neighbour. It’s very nuanced as a topic to see how we can be interpretable - it’s hard to see what is and what isn’t interpretable. Sara Hooker, Google Brain

  • We believe that in the future, regularization will allow you to detect labels and errors in data. We apply this term to adversarial networks. We found the best thing is to combine adversarial training with Lipschitz regularization, and we have the best published results to date. Adam Oberman, McGill University

  • We are seeing a massive drop in genome sequencing costs - in 2001 sequencing costs around $1mi currently it is now around $1000. By 2025 - human genomes storage would be around 40 exabytes, in comparison, YouTube’s storage is set to be around 2 exabytes in 2025. Helia Mohammadi, Microsoft

  • We’re not just trying to solve the problems that we have today, that we commonly see in the workload today. We are designing architectures that support the future of machine intelligence. Angel Serrano, Santander

  • We asked for input both inside and outside of government and received 26,000 ideas about how technology can help transform it. This is exciting - we have an opportunity in the digital age to transform the government. Hon. Peter Bethlenfavly, Treasury Board Secretariat of Ontario

  • Artificial intelligence is not standing still. We should not think the AI skill set needed in the future will be be same as the skill set we currently need in AI. How do we build and think about emerging and disruptive technologies in the interest of the society? Andrew Moore, Carnegie Mellon University

  • Developers have a responsibility to explain the technologies they are exploring. Policy makers need to understand the technology. These things are explainable and understandable and it is critical for policy makers to not interpret AI as a non understandable black box and will produce amazing results. Tegan Maharaj, MILA

  • What do these new technologies mean for government and social progress? We’re working to address these issues, and in policy we’re working to always maintain a level playing field. Ekkehard Ernst, International Labour Organization

  • It’s exciting to be here, I’ve met a lot of people who are visiting from outside Ontario - this is evolving as the global centre of AI - Ontario contributes more than 40% of Canada’s GDP across, FinTech, Healthcare, Business, and Ontario is proud of this. Joseph Kurian, Ministry of Economic Development


Women in AI Dinner

  • We make sure to test against diverse audiences, so we know what works and doesn't work for business owners from all backgrounds and experiences. It's incredibly easy to inadvertently introduce unintentional biases into machine learning algorithms, so all of us need to be active in checking our assumptions. Courtenay Siegfried, Alice

  • With our bot, Eno, we used natural language processing to classify and identify what the customer is saying so we can help them further. For us, this is a way that Eno is providing tangible value to our users. Margaret Mayer, Capital One

  • ‘bad data is any obsolete old data that’s inconsistent, not timely, irrelevant, one of the major issues I’ve seen is inconsistent data leading to inaccuracies. Giewee Hammond

  • ‘Our lab is directed towards research in understanding and implementing deep learning methods in the area of image segmentation and classification’. Rupa Kanchi, MD Anderson Cancer Center

Applied AI Summit
  • We're exploring models that actually make sense for automation - they lean towards the DL realm. When we explore bot automation, we look at this from a full automation context which is why we're using DL. Lynn Calvo, GM Financial

  • The main pain points in AI adoption are caused by barriers between AI and the rest of the business. Progress is slow without interaction between AI and the rest of the business. Biao (Bill) Chang, eBay

  • Imagine 10 years ago people would say 'if only I knew the internet was kicking off I would've bought nike.com' - the same thing is happening with AI today. if you have ideas, you should start developing them right now. Adam McMurchie, RBS

  • There's about a $65bn market opportunity in furniture, but people aren't comfortable to buy online. The key is emerging technologies and personalization to increase adoption. This creates trust and also a better UX than in store. Sunanda Parthasarathy, Wayfair

  • A small team that constantly develops better features and maintains a good ML model retraining pipeline will beat any army of analysts that manually handcrafts and manages hundreds of business rules. Hao Yi Ong, Lyft

  • We’re using ML to help lower the entry level into business. With Shopify capital, you join, start building your business, and find out how much money you’re eligible for. Then you accept and it’s straight in your account. Kyle Tate, Shopify

Machine Learning for DevOps Summit
  • I deal with all the issues you guys have. My specialist is around AI and big data, and I’ve now grown into DevOps. I’m the person who’s always been on a laptop writing your AI, and now I’ve finally moved to the cloud! I’m now right in the middle of ML and DevOps. Chandni Sharma, Google

  • DevOps has an incredibly diverse culture within it that is heterogeneous. It involves not just QA engineers, but IT specialists and even business people. It is my opinion that DevOps is a vision yet to be fully realised. Marios Fokaefs, Polytechnique de Montreal

  • The reality is that no person has any idea about the full extent of the possibilities of automation. We need to start trying to make sense of this in DevOps. Kohsuke Kawaguchi, CloudBees

  • DevOps is Inclusion, complexity, empathy, culture, automation, learning, measurement and sharing. I've been working on a new model that has internal agencies, that is essentially risk balanced and generative. Chris Corriere, SJ Technologies

...and that's a wrap!! If you're keen to catch a RE•WORK summit in 2019, check out our website and see which event best suits you. We'll be back in San Francisco, Boston, London, New York, Montreal and more, and we hope to see you there.  Original

Deep Learning Machine Learning AI RE.WORK News Deep Learning Summit AI Assistants


0 Comments

    As Featured In

    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original
    Original

    Partners & Attendees

    Intel.001
    Nvidia.001
    Acc1.001
    Rbc research.001
    Mit tech review.001
    Graphcoreai.001
    Twentybn.001
    Forbes.001
    Kd nuggets.001
    Ibm.001
    Facebook.001
    Maluuba 2017.001
    This website uses cookies to ensure you get the best experience. Learn more