Ayush Mangal

Ayush Mangal

Data & Applied Scientist

Microsoft, Hyderabad

Biography

Hi! I am a Data Scientist working in Bing Travel, helping make their visual content make people go WoW ❤️. I am currently interested in using Deep learning to make awesome products.

During my undergrad I had the pleasure of working in various academic research labs and startups on topics ranging from Computer Vision, to NLP, finance, reinforcement learning. I was also involved in leading various undergrad research groups, focusing on Deep Learning, Core CS, Quantum Computing and Blockchain.

Professional stuff aside, I am a die-hard anime fan, an avid reader, wannabe Musician/Writer/Youtuber? lol (checkout my YT channel and Medium!) and a lifelong learner who’s always up for a good conversation.

Interests
  • Computer Vision
  • NLP, MLOps
  • Content Creation
  • Anything and everything :P
Education

  • B.Tech. in Computer Science, 2022

    Indian Institute Of Technology Roorkee

Experience

 
 
 
 
 
Microsoft
Data & Applied Scientist
Microsoft
June 2022 – Present Hyderbad, India
Creating and maintaining various data-pipelines for visual content and developing novel methods to maintain image quality at scale.
 
 
 
 
 
Rephrase.ai
Research Intern
Rephrase.ai
December 2021 – January 2022 Bangalore, India
A deep‐tech startup that creates personalized videos for marketing using DL
 
 
 
 
 
RVL Lab, University Of Toronto
Research Intern
RVL Lab, University Of Toronto
August 2021 – May 2022 Prof. Florian Shkurti
Worked on Model based Deep Heirarchial RL
 
 
 
 
 
Machinge Learning Department, Carneighe Mellon University
Research Intern
Machinge Learning Department, Carneighe Mellon University
June 2020 – February 2021 Prof. Katerina Fragkiadaki
Worked on Unsupervised 3D Multi-Object Tracking
 
 
 
 
 
MIDAS Lab, IIITD
Research Intern
MIDAS Lab, IIITD
March 2020 – April 2021 Prof. Rajiv Ratn Shah
Worked on various multimodal NLP + Finance projects
 
 
 
 
 
VAL, IISC Bangalore
Research Intern
VAL, IISC Bangalore
December 2019 – January 2020 Prof. Venkatesh Babu
Worked on deep computational photography

Featured Publications

Modeling Financial Uncertainty with Multivariate Temporal Entropy‐based Curriculums (UAI 2021)
PDF Code Poster Video
Modeling Financial Uncertainty with Multivariate Temporal Entropy‐based Curriculums (UAI 2021)
Track, Check, Repeat: An EM Approach to Unsupervised Tracking (CVPR 2021)
PDF Code Poster Slides Video
Track, Check, Repeat: An EM Approach to Unsupervised Tracking (CVPR 2021)
Multimodal Multi‐Task Financial Risk Forecasting (ACM MM 2020 - Oral)
PDF Slides Video
Multimodal Multi‐Task Financial Risk Forecasting (ACM MM 2020 - Oral)
DEAP Cache: Deep Eviction Admission and Prefetching for Cache (Submitted to AAAI 2021)
PDF Code
DEAP Cache: Deep Eviction Admission and Prefetching for Cache (Submitted to AAAI 2021)
Visual Relationship Detection using Scene Graphs: A Survey (Submitted to ACM CSUR 2020
PDF
Visual Relationship Detection using Scene Graphs: A Survey (Submitted to ACM CSUR 2020

Projects

*
All Diffusers Paper Implementations LLM
Advanced RAG - FLARE

Advanced RAG - FLARE

A minimal implementation of an advanced RAG techinque - FLARE

Code
Tree of Thoughts

Tree of Thoughts

A minimal implementation of Tree Of Thoughts prompting

Code
DDPM Jax

DDPM Jax

Implemented a basic diffusion model in pure JAX

Code
DDPM Pytorch

DDPM Pytorch

Implemented a basic diffusion model in Pytorch

Code
Kandinsky 2.1, Diffusers

Kandinsky 2.1, Diffusers

Contributed to integrating Kandinsky 2.1 into Huggingface 🤗, Diffusers 🧨

Code
BLIP-Diffusion, Diffusers

BLIP-Diffusion, Diffusers

Contributed to integrating BLIP-Diffusion into Huggingface 🤗, Diffusers 🧨

Code
Consistency Models, Diffusers

Consistency Models, Diffusers

Contributed to integrating Consistency Models into Huggingface 🤗, Diffusers 🧨

Code
ADDA

ADDA

Implemented the paper Adversarial Discriminative Domain Adaptation in Pytorch

Code
Genzoo - VAE

Genzoo - VAE

Implemented a Variational AutoEncoder (VAE) in Pytorch

Code

Papers We Read

A collection of short summaries of papers we read

Code

Extra-Curriculars

 
 
 
 
 
Vision & Language Group, ACM IITR
Co-President
Vision & Language Group, ACM IITR
September 2019 – May 2022 IIT Roorkee
Deep Learning Research Group
 
 
 
 
 
RR With Deku
Youtuber
RR With Deku
October 2020 – Present Youtube
A Youtube Channel for discussing recent Deep Learning papers.
 
 
 
 
 
BlocSoc IITR
Founder
BlocSoc IITR
October 2021 – May 2022 IIT Roorkee
Blockchain & Web3 Group
 
 
 
 
 
ACM IITR
Chair
ACM IITR
March 2021 – May 2022 IIT Roorkee
Core CS Research Group
 
 
 
 
 
Quantum Computing Group, ACM IITR
Executive Member
Quantum Computing Group, ACM IITR
March 2021 – May 2022 IIT Roorkee
Quantum Computing Research Group

Recent Videos

.js-id-diffusion
Diffusion RL LLM
On the Expressivity of Markov Rewards

On the Expressivity of Markov Rewards

Evolving RL Algorithms

Evolving RL Algorithms

RL Using Diffusion Models

RL Using Diffusion Models

Textual Inversion

Textual Inversion

Dreambooth

Dreambooth

Flash Attention

Flash Attention

FrugalGPT

FrugalGPT

StreamingLLM

StreamingLLM

Prompt2prompt

Prompt2prompt

InstaFlow

InstaFlow

IntstructPix2Pix

IntstructPix2Pix

Contact