Untitled

Previously CEO and founder of MetaMind, a startup acquired by Salesforce, Richard Socher now leads the Salesforce’s research efforts and works on bringing state of the art artificial intelligence solutions to Salesforce. We spoke with Richard about his top priorities in advance of AI By The Bay and asked him to debunk the main myth about AI. 

What is the most exciting project you are working on right now as a Chief Scientist at Salesforce?

At Salesforce I wear two hats, one as Chief Scientist leading AI research and another in applied AI where I help bring state of the art aI technology to our customers with Salesforce Einstein. I’ll choose an exciting project from each.

On the AI research front, it’d have to be the “joint multi-task” learning model we recently introduced, a single deep neural network model which can learn five different NLP tasks and achieve state-of-the-art results. Our model starts from basic tasks and gradually moves to more complex tasks, which is continuously repeated until our model finishes learning all of the tasks. By doing this, our model allows the tasks to interact each other and improves accuracy.  

In terms of applied AI, with Einstein AI features and capability are embedded directly into the entire Salesforce platform and empower everyone– from engineers to customers– to leverage AI to be smarter, more productive and predictive. By doing the heavy lifting and removing the complexity of AI we are able to deliver seamless and scalable AI to Salesforce customers of all sizes.  

What are the key challenges for companies that are getting onboard with AI and machine learning?

Data is crucial for AI to work and most struggle with accessing historical data needed to build predictive models. Wrangling this data is expensive both in terms of time and resources and technical expertise is required to properly label outcomes for the AI model to predict. For most companies, the technical complexity and resources is just too much.

What are your priorities in terms of AI and machine learning at Salesforce going forward?

I’m focused on creating the world’s smartest CRM with Salesforce Einstein by advancing the science behind deep learning. This enables the creation of new applications of this technology.

The other priority is creating applications that thoroughly understand language, image and structured data in order to help businesses get answers to any question they have about their data and make more accurate and personalized predictions.

What is the biggest AI myth you would like to debunk?

There are so many floating around today, but I’d say the biggest myth is that AI will become independent, self-aware and take over. We are currently nowhere near anything resembling such a system, and these myths can be distracting from the actual, more pressing issues that could arise.

How can developers, data scientists, and software engineers engage with Salesforce?

Our platforms like Force.com and Heroku make it easy to build add-on apps that integrate into Salesforce’s main applications. We’re also continuing to add new services on top of these platforms that will empower anyone to build AI-powered apps.

You’ve spoken at Text By the Bay in 2015 and Data By the Bay 2016 on Deep Learning with NLP and computer vision. Where are we headed in 2017?

I think it’s time we build a larger, more versatile language and vision system that can do more than one thing, a.k.a multi-task learning.

Right now no one is able to train a single model to do many different kinds of things as once. For example, the question, “Who are my customers?” presents a simple listing task. “Who are my best customers?” adds a complex layer that requires numerous integrated tasks to answer many qualifying questions and make some hard decisions. With questions like the second one in mind, the research team at Salesforce recently created a “joint multi-task” learning model that successively trains with basic tasks, gradually moves to more-complex tasks, remembers all the tasks, and allows them to interact with each other. I’m really excited about it and I think we’ll see more on this front in 2017.

We’re excited to release the latest research from our machine intelligence R&D team! 

This report and prototype explore probabilistic programming, an emerging programming paradigm that makes it easier to construct and fit Bayesian inference models in code. It’s advanced statistics, simplified for data scientists looking to build models fast. Bayesian inference has been popular in scientific research for a long time. The statistical technique allows us to encode expert knowledge into a model by stating prior beliefs about what we think our data looks like. These prior beliefs are then updated in light of new data, providing not one prediction, but a full distribution of likely answers with baked-in confidence rates. This allows us to asses the risk of our decisions with more nuance. Bayesian methods lack widespread commercial use because they’re tough to implement. But probabilistic programming reduces what used to take months of thorny statistical sampling into an afternoon of work. This will further expand the utility of machine learning. Bayesian models aren’t black boxes, a criterion for regulated industries like healthcare. Unlike deep learning networks, they don’t require large, clean data sets or large amounts of GPU processing power to deliver results. And they bridge human knowledge with data, which may lead to breakthroughs in areas as diverse as anomaly detection and music analysis. 

Our work on probabilistic programming includes two prototypes and a report that teaches you:

How Bayesian inference works and where it’s useful

Why probabilistic programming is becoming possible now

When to use probabilistic programming and what the code looks like

What tools and languages exist today and how they compare

Which vendors offer probabilistic programming products

Finally, as in all our research, we predict where this technology is going, and applications for which it will be useful in the next couple of years.

Probabilistic Real Estate Prototype

One powerful feature of probabilistic programming is the ability to build hierarchical models, which allow us to group observations together and learn from their similarities. This is practical in contexts like user segmentation: individual users often shares tastes with other users of the same sex, age group, or location, and hierarchical models provide more accurate predictions about individuals by leveraging learnings from the group.

We explored using probabilistic programming for hierarchical models in our Probabilistic Real Estate prototype. This prototype predicts future real estate prices across the New York City boroughs. It enables you to input your budget (say $1.6 million) and shows you the probability of finding properties in that price range across different neighborhoods and future time periods.

Hierarchical models helped make predictions in neighborhoods with sparse pricing data. In our model, we declared that apartments are in neighborhoods and neighborhoods are in boroughs; on average, apartments in one neighborhood are more similar to others in the same location than elsewhere. By modeling this way, we could learn about the West Village not only from the West Village, but also from the East Village and Brooklyn. That means, with little data about the West Village, we could use data from the East Village to fill in the gaps! 

Many companies suffer from imperfect, incomplete data. These types of inferences can be invaluable to improve predictions based on real-world dependencies.

Play around with the prototype! You’ll see how the color gradients give you an intuitive sense for what probability distributions look like in practice.

How to Access our Reports & Prototypes

We’re offering our research on probabilistic programming in a few ways:

Single Report & Prototype (digital and physical copies)

Annual Research Subscription (access to all our research)

Subscription & Advising (research & time with our team)

Special Projects (dedicated help to build a great data product)

The start of a new year is an excellent occasion for audacious extrapolation. Based on 2016  developments, what do we expect for 2017?

This blog post covers five prominent trends: Deep Learning Beyond Cats, Chat Bots - Take Two, All the News In The World - Turning Text Into Action, The Proliferation of Data Roles, and What Are You Doing to My Data? 

By Dan Pelleg, Yuval Pinter, and David Carmel

There’s no denying it: chat bots are en vogue and it seems like everyone is experimenting with the technology. Facebook introduced bots for their Messenger Platform, Microsoft launched a controversial chatbot on Twitter called Tay, and at Yahoo, we’ve released our own bots on Kik and Messenger. With so much interest, we’re doing our part as research scientists to advance the state-of-the-art in question-answering, which, among other things, will lead to conversational bots appearing more human.

For the second year in a row, the Yahoo Research Text Mining team in Haifa, in collaboration with Emory University and the U.S. National Institute of Standards and Technology, ran a shared challenge for question-answering called LiveQA as part of the annual Text REtrieval Conference (TREC).

During the 24-hour duration of the competition, 14 teams from the USA, China, Germany, Australia, Canada, Israel, and Qatar participated, though preparations for the competition actually started a few months earlier. Each team developed a Web service that, as input, received a free-form question, and then responded with an answer. The answer could not be longer than 1000 characters, and needed to be returned within one minute. The questions used, from Yahoo Answers, were different than the factoid questions used in some previous TREC QA tracks. Yahoo Answers questions are much more diverse and cover multiple question types, such as opinion, advice, and polls. This made the task far more realistic and challenging.

During the competition day, the TREC participants received a total of 1,088 questions and sent back a total of over 21,000 answers. They usually had no issue with either the time limit (24 seconds to answer, on average) or the length limit (599 characters per average answer).

The answers were judged manually on a four-point scale (excellent/good/fair/poor). The best system in terms of performance was the “EmoryCrowd” system from Emory. This was a unique system, because it used a combination of computational and human labor. First, it computed candidate answers algorithmically, and then it turned to crowdsourced work to rank them – all in less than 60 seconds. This “cyborg” system provided the highest-quality answers by a noticeable margin, trailed by fully-algorithmic systems from CMU, Emory, and Yahoo Research.

Human intellect was shown to be superior not just within hybrid systems, but also by itself, when we had the judges also evaluate the answers given by users on the original Yahoo Answers site. These answers were significantly better than any automated or hybrid system. Of importance though, the Yahoo Answers users had no time or space limit – if they wanted to, they could work on their answers for up to one week, which would give them an obvious advantage. This is just one reason to keep pushing the envelope on automatic question answering. Computers are faster than people for many tasks, they are available 24/7 (if the power is on), and they scale more readily.

We are proud to introduce ‘social media bookmarking’, the newest integration into Diigo.

This is the age of social media platforms. The way we consume information from the internet has evolved, and so has Diigo.

Just about everyone is on a couple social media platforms, whether to consume or share online content. Managing all your favorite pieces of information throughout all these platforms and apps becomes very difficult once you are using a few sites together. In the case of heavy social media users, the task of managing all this content becomes overwhelming.

Social media bookmarking from Diigo helps you store, find and organize all your favorite pieces of content from these platforms all in one place; your Diigo library.

Connect the platform or your choice to your Diigo account, and your ‘likes’ and ‘favorites’ will automatically be sent to your Diigo library. Default tags can be added to saved content from each separate platform, or you can add/edit them manually from your library. From your library, you can also easily add them to Outliners for further organization, or share them to your Groups directly.

Setup a free Diigo account and give social media bookmarking a try! All Diigo accounts can connect one social media platform for free, while upgraded accounts can connect all the platforms we support simultaneously.

https://www.diigo.com/aggregation

Many more platforms are planned to be added for Diigo integration. Check out our currently supported platforms below!

Twitter - ‘Favorite’ a tweet to send it to your library

Pocket - Choose between ‘favorite’, specific tag/s or ‘all’ for what to send to your library

Instapaper - Choose between ‘favorite’, a specific folder, or ‘all’ for what to send to your library

Github - ‘Star’ a repository to send it to your library

Stack Overflow - ‘Favorite’ a question to send it to your library

Youtube - ‘Like’ a video to send it to your library