About Me
I'm a Master's student in Computer Science at University of Massachussets, Amherst. My Research interests are primarily in Reinforcement Learning, Causal Inference and Deep Learning. Prior to this, I was working as a Machine Learning Engineer at LinkedIn on content quality .
Personal Information
- Name : Saket Tiwari
- Phone : +1-630-550-6016
- Email : sakett786@gmail.com
- City : Amherst, MA, USA
Reinforcement Learning
Natural Option Critic
Derived and demonstrated advantages of a natural gradient based learning algorithm. Worked on Deep Hierarchical Reinforcement Learning using theano. We published theorems specifying a tractable form of the Fisher Information Matrix. This was for the parametrized options framework introduced in option critic framework.Saket Tiwari, Philip S. Thomas; Accepted to AAAI 2019 main track, selected for spotlight presentation.
Hyperbolic Embeddings for Learning Options in Hierarchical Reinforcement Learning
Developed a novel method to split Reinforcement Learning tasks into meaningful sub-tasks. Applied the topology of hyperbolic spaces to learn embeddings. Inspired by the recently proposed Poincare embeddings . We obtained improved results for two toy environments. We plan on extending it to more complex tasks.Saket Tiwari, M Prannoy; Under double blind review at AISTATS.
Deep Learning and Causal Learning
Cache Miss Rate Predictability via Neural Networks
Designed a Deep Neural Network for predicting cache miss rates in benchmark programs, inspired by WaveNet model for audio generation. Released a new data set of cache miss rate "signals". First of its kind approach to studying cache miss rates using Machine Learning.Rishikesh Jha*, Arjun Karuvally*, Saket Tiwari*, Eliot Moss; Accepted for poster presentation at NeurIPS 2018 workshop on ML for Systems.
Causal Effect Inference in Social Networks via Macro-Variable Manipulation
Developed a causal framework using Variational Auto-Encoders for social networks, applying macro-variable manipulation. We answer the counterfactual question "what if the user had behaved this way?", then could they have affected the outcome positively?Saket Tiwari, Matthew Rattigan; Submitted to AAAI Spring Symposium on Beyond Curve Fitting.
Non Refereed
Memory Access Entropy Prediction
Developed a model for predicting entropy values of memory address accesses. Designed and developed a Deep Neural Network fashioned after WaveNet to model entropy behavior.Saket Tiwari, submitted in completion of Independent Study under Professor eliot Moss.
Understanding the Cache Features in Intel Core Processors
Reverse engineered cache pre-fetching features in modern intel core processors: i3, i5 and i7. Devised experiments to demonstrate the posited cache pre-fetching mechanisms. Experiments directed towards devising a cache based side channel attck in intel core processors.Saket Tiwari, Ashok T, Pratik Kumar; Submitted in completion of B Tech Project at IIT Bombay under Professor Bernard Menezes.
Education
I have recieved the following scholarships in the past:
- ⇒ KVPY Scholarship. Offered to young scientists by the government of India. 2008-09
- ⇒ Sri Chaitanya Scholarship. Offered for outstanding performance in academic in the form of school fee waiver. 2008-10
2017 - 2019
Studying for completion of Master of Science in Computer Science. My primary research interest is Reinforcement Learning. My Master's project is on Causal Learning, where I am applying it for the LinkedIn Economic Graph Research Challenge
2010 - 2014
Recieved a Bachelor of Technology, with honors, in Computer Science. BTech Project was on reverse engineering cache pre-fetching behavior in modern Intel processors: i3, i5, and i7. Participated in various extra-curriculur activites and was responsible for upholding the Hostel constitution.