What Are Deep Learning Nets? Define Mathematical models that mirror human neurons Recognizing patterns like images, sounds, verbal sequences Neural nets need those patterns to be translated into numbers Adam Gibson Skymind
Who Is Deep Learning For? Users Chief Information Officers looking for the highest performance in machine learning Businesses looking to maximize the value of small data science teams, creating apps and cutting costs. Data scientists and programmers who seek to process huge amounts of unstructured data without laborious feature extraction. Adam Gibson Skymind
Brands
"The biggest disruptor that we are sure about is the arrival of big data and machine intelligence. This disruption will not only change every business globally, it will also have an important impact on the consumer." Google Chairman Eric Schmidt, Bloomberg TV, Dec. 30 2013
First Movers Google hired Geoff Hinton, the pre-eminent deep learning expert, last year to help build their search and ad recommendation engines. Facebook hired Yann LeCun, the no. 2 deep-learning expert, to help build a Facebook feeds that recommend more engaging content. Baidu hired Andrew Ng, a Stanford deep-learning professor, away from Google to lead its research team. Microsoft, Amazon and Netflix all have deep-learning teams working to give customers more of what they want through recommendation and search. Tech giants have monopolized deep learning.
A breakthrough in machine learning would be worth 10 Microsofts. - Bill Gates, Microsoft Machine learning is the next Internet. - Tony Tether, DARPA Director Web rankings today are mostly a matter of machine learning. - Prabhakar Raghavan, Yahoo Director of Research Machine learning is going to result in a real revolution. - Greg Papadopoulos, Sun CTO Machine learning is todayʼs discontinuity. - Jerry Yang, Yahoo Deep learning is that breakthrough, and Skymind is bringing it to industry.
Domains IMAGE FACES MACHINE VISION SOUND SPEECH TO TEXT MACHINE TRANSLATION TEXT SEARCH INFORMATION RETRIEVAL TIME SERIES
Faces Learned By Our DL Networks Adam Gibson Skymind
Numbers Grouped by Skymind
Use Case Search => Google, Bing, Yahoo Information retrieval Document clustering Text, sound and image search Question-Answer ~ Watson
Use Case CRM Customer Resource Management Churn prediction (SaaS) Big spender prediction (gambling) Log/behavior analysis Sentiment Analysis
Use Case Recommendation Engine E-commerce => Amazon Advertising => Google, Baidu See: "The Chinese Google Is Making Big Bucks Using AI to Target Ads" --Wired, Oct. 20, 2014
Use Case Enterprise Resource Management/Planning Factories Complex Logistics Personnel Deployments Oil & Gas
Background
Cloud Big data lives in the cloud Nets train on big data for good results Training = computationally intensive Deep learning also lives in the cloud
Deep learning... Signal & Scale Needs scale to be powerful Can identify signal at scale Achieves scale with raw data And thousands of CPUs or GPUs Running in parallel
Deep Learning Is Setting Records Deep Learning Difference Accuracy Machine Learning Amount of Data
Lessons The world is more knowable than ever before. Machines can measure events and human behavior more accurately. Business has not caught up with its own data. Machines can grow smarter with deep learning. Analyzing huge seas of data that no human has ever seen. Knowledge is only power... if you interpret data.
Method Deep learning classifies the world. It starts with obvious things (which are not obvious for machines), and extends to subtle things, which are not obvious for humans. Nothing is obvious to machines because they only understand numbers. Houses, men, dogs, cats, trees are all numbers.
Why Is Deep Learning Hard? We see this Machines see this Problem (Hat tip to Andrew Ng)
Take aways Be careful out there.
Contact Adam Gibson Skymind Founder adam@skymind.io