Research/Blog
MEETING MINUTES FROM SATURDAY 25TH. MAY – AI LAB SESSION IN BENGALURU
- May 28, 2019
- Posted by: vsinghal
- Category: Deep Learning Machine Learning Reinforcement Learning
#CellStratAILab #disrupt4.0 #WeCreateAISuperstars
We had another round of deep AI sessions last Saturday in BLR AI Lab meetup.
![](http://www.cellstrat.com/wp-content/uploads/2019/05/Collage-1024x768.jpg)
I started with a detailed deep-dive on the Maximum Likelihood Estimation (MLE) algorithm for Logistic Regression, which is a type of classification technique. Logistic Regression has a concave optimization curve wherein we try to maximize the Likelihood of the positive class. By adjusting weights over many iterations, one can reach the peak of the concave likelihood curve where ideal weights are found. I also demoed use of Logistic Regression for classification of the IRIS flower dataset.
Next came a superb presentation on detecting blood cell types from blood slide images by Ramapriya. Blood Cells can be of 4 types : Lymphocytes, Monocytes, Neutrophils and Eosinophils. Also Nucleus Type can be mononuclear or polynuclear. Using a Keras-based Deep CNN network, one can detect the blood cell type. The model accuracy is checked with help of the accuracy score, precision and recall metrices. A model training over 40 epochs gives an accuracy of 93%.
After this, Shubha M (team lead for Finance Svcs AI research group) presented a fabulous session on Deep Q Learning, one of the most popular techniques for Reinforcement Learning. Shubha started with explanation of Q-value iteration algorithm as well as Bellman’s Equation. Deep Q Learning involves Q values update in an iterative fashion based on maximising current and future rewards potential in each state. Also at play is Exploration vs Exploitation, the former means exploring new paradigms by choosing random actions, whereas the latter involves going for the most optimized action (safe bet) in each state. A model starts with high Exploration quotient (to attempt to discover better rewards) but slowly settles for more Exploitation actions eventually.
Shubha demonstrated how DQN (Deep Q Net) can be used to trained to learn to play the video game Doom. A game video has a set of image frames that can be used to train a DQN model in order to learn to play the game.
Join us for the first AI Lab Hackathon on Saturday 8th June 2019 at BLR :-
Register : https://www.meetup.com/Disrupt-4-0/events/jvfhvqyzjblb/
Hackathon Topic : Object Detection and Localization in Images
Date : Saturday 8th June 2019, 10:30AM – 5:00PM
Loc. : WeWork, Embassy Tech Village, ORR, BLR
Sharpen your knives..err..computers and go for the kill..oops..AI models and solve the hackathon challenge. See you on 8th June for the AI Lab hackathon !
PS : Amazon Gift Certificates for the hackathon winners !
Questions ? Call me at +91-9742800566 !
Best Regards,
Vivek Singhal
Co-Founder & Chief Data Scientist, CellStrat
+91-9742800566