Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
Do you really want to delete this prezi?
Neither you, nor the coeditors you shared it with will be able to recover it again.
Make your likes visible on Facebook?
You can change this under Settings & Account at any time.
Transcript of Cognitive Computing
VP Core Technology
IBM Watson is the new face of IBM
New class of Technologies
New class of problems
Silicon Alley HQ
Start Up Culture
Hiring Best and Brightest
Statistical Machine Translation (IBM 1993)
Help diagnose cancer
Statistical Speech Recognition (IBM 1976)
One day one of his linguists resigned, and Fred [Jelinek] decided to replace him not by another linguist but by an engineer. A little while later, Fred noticed that the performance of his system improved significantly. So he encouraged another linguist to find alternative employment, and sure enough performance improved again. The rest as they say is history, eventually all the linguists were replaced by engineers (and not just in Fred’s lab) and then speech recognition really started to make progress.
Steve Young (2010),
Frederick Jelinek 1932 – 2010 : The Pioneer of Speech Recognition Technology
132 R&D openings for my team in NLP, ML, and cloud for 2014 alone.
Re-invent customer service
Speech to Speech
Open problems that involve perception, intuition, reasoning, and learning that only humans can solve.
Need for a new methodology
What It Takes to compete against Top Human Jeopardy! Players
Our Analysis Reveals the Winner’s Cloud
The Jeopardy Challenge
Huge commitment made in 2007 to bridge the gap!
Early choice: it's going to be messy
We can't use one large knowledge model.
No silver bullet algorithm.
Need to combine 100s of heuristics.
Need to combine 100s of information sources as-is.
Combine IR, NLP, ML, KRR, DA into a distributed architecture
Rapid Innovation Methodology
Systems Neuroscience and Machine Learning
"What I cannot build, I cannot understand"
"What I can build, I can no longer understand"
DeepFace - 97.25% accuracy
Breaks Captcha with 99.8% accuracy
Read Lips with same accuracy improvement as humans
Who am I?
Ideally experiments would be unit tested like any incremental change but it's much more complicated:
Need to run on realistic data at scale.
Need to use blind sets - testing too much reveals the sets.
Isolation is anyone's best guess.
What are bugs when there is no such thing as an expected behavior?
Evaluate potential cost/benefit before implementation.
Research existing literature.
Create experiments pipeline.
8000 experiments in 4 years. Less than 20% made it in.
Can you productize that methodology?
No ground truth readily available.
Each domain has a different version of the truth.
In many domains there is no best answer.
How do you do continuous integration when validation is so expensive?
Can you tell your boss you are throwing away 80% of what you write?
Use robust NLP and ML algorithms (e.g., deep learning)
Data, data, data - be creative in collecting good data (e.g., crowdsourcing, mechanical turk)
Provide great tools for domain adaptation and evaluation
Hire people like you.
News ways of developing software for a new era of computing