In this Lunch & Learn session, Chirag Jain gives us a friendly & gentle introduction to Machine Learning & walks through High-Level Learning frameworks using Linear Classifiers.
2. About Haptik
Chatbot platform for publishers, advertisers and enterprises
AI powered conversational interface to drive customer engagement
Reach of 30 Million Users, processing 5 Million Chats per month
One of the worldâs largest chatbot platforms
Started in 2013, global pioneers of chatbots
3. How this talk is divided
Part 1: AI Introduction and applications
âIntroduction
âNew and Old news about AI
Part 2: ML Introduction and workflow
âIntroduction
Part 3: High Level Learning framework
âCode (and some Math) walkthrough of linear
classifier
4. What is AI ?
Demonstration of human like intelligence by
machines.
A machine performing any task that needs human
level intelligence can be said to be âArtificially
Intelligentâ
5. AI In everyday life today
Email Categorization Web search
Targeted (annoying) Ads
6. AI In everyday life today
Maps & Navigation Computer Games Digital Assistants
9. Few ML success stories in the past 3 years
Neural Style Transfer Controllable Image Generation
(Xianxu Hou et. al.)
10. Major Goals of AI
âReasoning and Problem Solving
âKnowledge Representation
âAutonomy and Planning
âSelf Learning via Experiences â Machine Learning is a part of this
âNatural language processing
âSensory Perception
11. Major Goals of AI
âMotion and Manipulation
âSocial Intelligence
âGeneral/Super Intelligence â Media tries to sell you this
12. Sciences involved in AI research
âComputer Science
âMathematics
âPsychology
âLinguistics
âPhilosophy
âMany Others
13. Philosophy around AI
âIs general/super intelligence possible ?
âDo they have to be similar to human systems to be
intelligent as us ?
âCan intelligent machines be dangerous ?
âShould we prefer more accurate systems over
transparent systems ?
14. The vagueness and the hype
Real Story: Task was to learn negotiation in natural language, not
some efficient cryptic language. Researchers only reported a failed
experiment trail.
15. Few words on âDeep learningâ
Image Credits: http://neuralnetworksanddeeplearning.com/chap5.html
16. AI, ML, NN, DL are not new!
âFirst Programmable Computer â 1936
âAI research began â 1956
âNeural Networks - base ideas as early as 1943, polished idea â
1958, research active since 1990s
âDeep learning - first idea proposed in 1965, early
implementations â 1965 - 1971, research active since 1990s
âLarge NNs were computationally infeasible to train back then
âNNs and DL went into âhibernationâ for more than a decade
17. Resurgence of âAIâ because of Deep Learning
Training complex models has become feasible now
âLarge datasets are available for some tasks
âCompute power has increased exponentially - we now
have very powerful GPUs/TPUs
âTheoretical ideas in research have been polished over
time
âMuch better tools to work with!
â Theano,Tensorflow (Google), Keras (now Google), Torch/PyTorch(Facebook), CNTK
(Microsoft), Caffe (UCB), MXNet(Apache, Amazon), sklearn, gensim, nltk
19. Machine Learning
Blends ideas from statistics, computer science, operations
research, pattern recognition, information theory, control
theory and many other disciplines to design algorithms
that find low-level patterns in data, make predictions and
help make decisions (at scale).
21. Common Taxonomy of ML methods
âSupervised Learning - some feedback is available
âCompletely Supervised Learning
âSemi-Supervised Learning
âActive learning
âReinforcement Learning
âUnsupervised Learning - no explicit ground truths
âMeta learning
â...
22. Common Tasks for ML
âClassification (usually supervised)
âRegression (usually supervised)
âClustering (unsupervised)
âDimensionality Reduction
â...
23. Classification
âTask is to learn to categorize input into discrete classes
E.g. Input: Image
Output: probabilities of image containing
{dog, cat, horse, zebra}
âSupervised task, we have true labels for each input
âMetrics: To keep things simple, we will use accuracy -
how many things the classifier can classify correctly.
Selecting a metric depends on the data + problem
24. Logistic Regression - A simple linear classifier
Notebook to follow along
https://gist.github.com/chiragjn/24b548785d99a393fca9dc
cfe1439d4a
33. Other things that we donât have time for
âNon-Linear classifiers
âLearning Methods that donât use Gradient Descent
âOther Metrics: Precision, Recall, F1
âOverfitting and underfitting
âAnd many more tricks of the trade