Sai Aditya Chitturu

Sai Aditya Chitturu

Machine Learning Researcher


About Me

Greetings! Yours truly is a graduate student at University of California, San Diego (UCSD), pursuing his Masters degree in Computer Science (with a focus on Machine Learning). My main research interests revolve around Deep Learning, General Intelligence and Computer Vision :)

Since the past few months, I got to work on fun stuff (Action recognition for Smart homes, SMART Natural Products Identification, etc.) as a Graduate Student Researcher at The Cottrell Lab (GURU).

I am actively seeking out full-time opportunities in Machine Learning and am also looking for collaborations on cool projects / hackathons. Please don't hesitate to reach out.

To make this more about me, here's what I've been doing over the couple of years.

Work Experience

Graduate Teaching Assistant - CSE @ UCSD
(Jan 2018 - present)

  • Assisting with the course "Neural networks" for Computer Science graduate students (~400 students)
  • Held tutorials on PCA, LSTM, PyTorch, etc. and helping students with their projects and understanding
[Python, PyTorch, NumPy, TensorFlow]

Graduate Teaching Assistant - CSE @ UCSD
(Oct 2018 - Dec 2018)

  • Assisted with the course "Recommender Systems" for Computer Science graduate students (~410)
[Python, NumPy, Pandas, TensorFlow, scikit-learn]

Graduate Student Research Assistant - The Cottrell Lab (GURU)
(Jun 2018 - present)

  • Working on Action Recognition for a SMART Home project and another project(SMART) aimed at speeding up natural parts discovery using deep learning
[Python, NumPy, PyTorch, Keras, C++]

Graduate Teaching Assistant - Rady School Of Management
(Apr 2018 - Jun 2018)

  • Assisted with the course "Data Science for Finance Using Python" for Senior year management students
  • Held tutorials on Pandas, NumPy, SVM, etc. and aided student with projects and kaggle competitions
[Python, NumPy, Pandas, scikit-learn]

Application Developer - JPMorgan Chase & Co. (JPMC)
(Jul 2014 - Jul 2017)

  • Worked in a core team of 6 to develop and test a resilient web application in an agile manner, using JAVA Spring MVC, AngularJS, Bootstrap and Oracle DB
  • Mentored interns and a University Outreach Program(UOP) team at JPMC. Runner-up Team in UOP contest 2015
  • Developed components and functionalities for Claims suite of applications, which handle Debit Card claims for the firm
  • Identified and implemented changes that eventually reduced the operations time by 10 min a day / operations engineer
[Java, JavaScript, AngularJS, SQL, Eclipse, Git, SVN, JIRA, ALM, Spring MVC]

Research Analyst Intern - NetApp
(Jan 2014 - Jun 2014)

  • Worked at Advanced Technologies Group on “In Memory Hot and Cold Data Tagging”
  • Implemented data mining techniques to determine spatial and temporal re-access patterns
  • Implemented an ensemble of scheduling algorithms aimed at obtaining higher hit rates and a better caching technique when in-memory databases are used for OLTP. Arrived at a model with a better hit-rate than the LRU standard
View Report
[Python, AWK, Splunk]


Action Recognition for Smart Homes

  • Detecting and classifying actions based on skeleton joint locations. Employed echo state networks to identify the transition of an action to another
  • Employed existing architectures including a ResNet, a CNN + LSTM model, Echo State Networks and Graph Convolutional Network
  • Currently working on novel architecture to recognize actions capable of running on a Raspberry Pi

Layerwise (Divergence) Analysis of Adversarial Perturbations

  • Performed a causality analysis and layer-wise divergence analysis of the adversarial perturbations on a deep network.
  • Explored the characteristics of the adversarial perturbations and thier effect on various layers of standard neural network architectures.
  • Tested defense strategies (preprocessing, entry network) which can aid in making the network more robust.
View Report

Music Generation using RNN Architectures

  • Experimented with different architectures of Recurrent Neural Networks to generate music at a character level using ABC notation.
  • Utilized LSTM and GRU architectures with varying levels of hidden units and layers to get the best performing model which could generate respectable music.
View Report

Prediction of Salary Income Ranges using Supervised Learning

  • Built a predictive model that predicts the possible salary range of a user based on a few selected features.
  • We used the Stack Overflow 2016 data-set which provides a myriad of features for the said task. Used various Supervised Classification Learning Algorithms like Random Forests, XGBoost, LightGBM, Neural Networks and compared the performances.
View Report

Image Captioning using Deep Neural Networks

  • Developed a novel architecture where an additional RNN layer was introduced in between the CNN and LSTM layers for image captioning.
  • The advantages and disadvantages of the new architecture were analyzed in detail using metrics like BLEU score. Further details in the report.
View Report