Divyam Srivastava

I got acquainted with AI/ML stuff in my freshman year and ever since December of 2017, I am undertaking various projects on Learning algorithms, Computer Vision, & NLP. So Far I have my experience with many data science libraries and have quite a good grip on frameworks like Keras and PyTorch. My research goals are now shifted towards attention-based scene understanding. Currently, we are witnessing a transition from traditional RNNs that are based on immediate one-directional attention, to transformers (with global attention mechanisms). Such attention-based models can open interesting avenues related to understanding how the human brain uses past experiences to model its current perception and selective focus abilities.

**Machine Learning Intern**

May 2019 - July 2019 · Niramai Health Analytix Pvt. Ltd. (Bangalore,India)

• Paper accepted at San Antonio Breast Cancer Symposium(SABCS)‘19
• Mammographic Assessment for cancer prediction provides inconclusive results for Dense Breasts, adding an unnecessary exposure to radiation
• Applied SOTA Deep Learning Techniques on thermographic breast images for density estimation (based on BIRADS category) as a pre-mammo process
• Optimized existing models by leveraging transfer learning techniques and custom optimizers with differential learning rates across all layers for stable training
• Analysed the effects of various views (per subject) as input on the final estimation
• Implemented a cascaded network for multi-view multi-input analysis as well
• Achieved a 90/60 specificity/ sensitivity ratio for dense breasts using ensemble of all proposed techniques.

B.Tech in Electronics and Communication Engineering

2017 - 2021 · Sardar Vallabhbhai National Institute of Technology, Surat

GPA(%): 8.60

Intermediate

2016 · Chanderbala Modi Academy

GPA(%): 90.6

Matriculation

2014 · Chanderbala Modi Academy

GPA(%): 10.0

Symposiums

2019. D. Srivastava, A. Ganesh, S. Kakileti, G. Manjunath. A non-radiation based screening to detect dense breasts. [San Antonio Breast Cancer Symposium, AACR].

Image Captioning

Exploring supervised techniques involving treating the problem as seq2seq task and thereby using Transformers to solve problem.

iHack 2019 Project - IITB (Winner)

Web App and Android App with a feature for flagging bad comments on a live video stream. Comments are automatically analysed and flagged using a deep learning model designed from scratch.

GAN paper Implementations

Implemented Traditional GAN, DCGAN, WGAN (Weight clipping), WGAN (Gradient Penalty) on MNIST, CelebA datasets.

Twitter Sentiment Analysis (Char level + Word Level CNN)

Experimented with embedding layers and added 1-D Convolution layers on word vectors; networks inspired by AlexNet, ResNet. Best Accuracy ~ 83.5%

Student Intervention System (ML + j-Fuzzy) | DotSlash 2019

Used Machine Learning models coupled with a j-Fuzzy Logic model to predict a student’s end semester marks based on various socio-economic factors.

Speaker Recognition in Attendance | DotSlash 2018

Used Machine Learning API from Microsoft Azure Cloud ML services to create a speaker recognition pipeline for attendance system in colleges.

• Reading Novels
• Redditing
• Blogging