I give an overview of current state of natural language analysis using machine learning algorithms. #naturallanguage
#machinelearning #artificianintelligence
2. Bio
• AI researcher published 10
research papers on topics like
logical reasoning, language
analysis, Semantic web.
• Masters from University of
Cambridge, UK.
• Involved with startups since
2010.
• Machine learning enthusiast.
3. What is machine learning?
• Apply previously acquired
knowledge to new or novel
situation
• Search tree, neural network,
Bayesian reasoning, logic,…
• Boom and AI winter cycle
(1974-80, 1980-87)
Arthur Samuel with his checker playing machine
4. But something is happening
recently…..
• AlphaGo defeated world Go
champion.
• AP is going to use machine
created news articles for
sports coverage
• Deep Mind to check NHS eye
scans for disease analysis
• People have termed it similar
to Industrial revolution
5. Applications
• News articles (AP), robot lawyers, designers
(wix), car industry (google, apple), tour guides,
rockets….
• Open AI, facebook, google, microsoft, twitter…
• ….machine learning will affect all domains…..
6. Why now?
• Big data - What do we do with
it?
• Visualize, Analyze - Human
element
• Machine learning / Deep
neural network - Learn from
the data
7. Language understanding
• “A computer would deserve to
be called intelligent if it could
deceive a human into
believing that it is human.” -
Alan Turing
• Language is the form of
communication.
• Basic necessity in solving AI
problems in language
understanding.
8. Applications of NLP
• Topic modelling
• Text summarization
• Translation
• Sentiment analysis
• Image captions and descriptions …
9. Historic approaches…
• Syntax tree
• Semantic - RDF, OWL
• LSA - bag of words,
similar words appear
together.
Concepts are represented
as patterns of words.
10. Latent Dirichlet Analysis
• Start with document, bag of words and K topics
• Output - Documents are of what topics (in %)
• Randomly/semi-randomly assign each word to a topic
• All topic assignments except for the current word in
question are correct
• Improve by reassign ‘w’ a new topic. Choose topic t
with probability p(topic t | document d) * p(word w |
topic t)
22. AlphaGo - Building intuition
• Took 150,000 games played by good
human players and used an artificial
neural network to find patterns
• Learned to predict with high
probability what move a human
player would take
• Play against itself, to get an estimate
of how likely a given board position
was to be a winning one -
Reinforcement learning
• No detailed knowledge of Go.
Instead analyzed thousands of prior
games and engaged in a lot of self-
play.