Avatar

Keshaw Singh

Member of Technical Staff II

Adobe Inc.

Biography

I am currently working in the algorithms team of Adobe Advertising Cloud Search (prev. Adobe Media Optimizer), where I build models which manage ad spends for businesses. I graduated from IIT Kanpur in the summer of 2017, with a major in Mathematics & Scientific Computing and a minor in Artificial Intelligence.
I am fascinated by techniques that can help generate information in human languages, to accomplish useful tasks like machine translation and abstractive summarization. In particular, I am intrigued by methods based on the transfer learning paradigm, as I believe they can make up for the paucity of training data in low-resource languages.
I am an avid football fan and spend most of my weekends watching matches. Sometimes though, I try to brush up my Spanish.

Interests

  • Neural Text Generation
  • Natural Language Processing
  • Transfer Learning

Education

  • BS in Mathematics & Scientific Computing, minor in Artificial Intelligence, 2017

    IIT Kanpur

Experience

 
 
 
 
 

MTS II

Adobe Inc.

Feb 2019 – Present Bengaluru
  • Implemented clustering models for ad messages using LDA as well as lexical matching
 
 
 
 
 

MTS

Adobe Inc.

Jul 2017 – Jan 2019 Bengaluru
  • Prototyped a neural network-based collaborative filtering for recommender systems which outperformed matrix factorization approaches by a large margin
  • Built a RetinaNet-based model for click-to-action (CTA) button detection in creatives
 
 
 
 
 

Product Intern

Adobe Media Optimizer

May 2016 – Jul 2016 Bengaluru
  • Twitter-driven improvement for AMO Click/Intraday model

Projects

*

Transfer Learning in Deep RL

Evaluate using policies learned from harder tasks.

Low-resource Neural Machine Translation

NMT with transfer learning and MUSE.

Microsoft AI Challenge

Answer selection given a search query.

Stochastic Variational Inference

Poisson Matrix Factorization with stochastic updates.

Query-based Summarized Email Extraction

Extractive summaries over Enron email dataset.

Fine-grained Vehicle Classification

Real time object classification in video.

Contact