Jyoti Aneja

Jyoti Aneja is a Member of Technical Staff on the Microsoft Research AI Frontiers team and a core contributor to Microsoft's Phi family of small language models. Trained in physics at IIT Kanpur and the University of Illinois at Urbana-Champaign, they completed a PhD at Illinois in 2021 with Alex Schwing, working on image captioning and variational autoencoders. After joining Microsoft in November 2021, they coauthored Textbooks Are All You Need, which introduced phi-1 in 2023, and the Phi-3 and Phi-4 technical reports in 2024. In March 2026, they co-led Phi-4-reasoning-vision-15B, an open-weight multimodal reasoning model for math, science and user-interface tasks.

Now at: Microsoft Research
Badges: Bridge

Timeline

2012–2015

Physics degrees in Kanpur and Illinois

They earn an MS in physics at the Indian Institute of Technology Kanpur in 2012 and a second MS in physics at the University of Illinois at Urbana-Champaign in 2015.
2018

Convolutional image captioning

As a PhD student working with Alex Schwing at Illinois, they coauthor Convolutional Image Captioning with Aditya Deshpande and Schwing at CVPR 2018. The paper develops a convolutional technique for writing image captions and tests it on the MSCOCO dataset.
2019

Diverse captions at ICCV and CVPR

At ICCV 2019, they present Seq-CVAE with Harsh Agrawal, Dhruv Batra and Schwing, a captioning model that learns a latent space for every word. A CVPR 2019 paper with Deshpande, Liwei Wang, Schwing and David Forsyth first predicts a part-of-speech summary of an image, then generates the caption from it.
2021

A PhD and contrastive VAE priors

They defend their PhD thesis at Illinois in September 2021. With Schwing, Jan Kautz and Arash Vahdat, they publish a NeurIPS 2021 paper on noise contrastive priors, an energy-based prior trained by noise contrastive estimation that improves the generative performance of variational autoencoders.
November 2021

Joining Microsoft Research

They join Microsoft in November 2021 and work as a senior researcher in the Deep Learning Group at Microsoft Research, Redmond, on problems that combine computer vision, deep learning and natural language processing.
2023

Phi-1 and Phi-2

In June 2023, they coauthor Textbooks Are All You Need, which introduces phi-1, a 1.3-billion-parameter language model for code trained on textbook-quality web data and synthetic textbooks. Phi-2, a 2.7-billion-parameter small language model from Microsoft Research's Machine Learning Foundations team, follows in December 2023 with Aneja among its contributors.
2024

The Phi-3 and Phi-4 reports

In April 2024, they coauthor the Phi-3 technical report, which introduces phi-3-mini, a 3.8-billion-parameter model small enough to run on a phone. In December 2024, they are a core contributor to Phi-4, a 14-billion-parameter model whose training recipe centers on data quality and synthetic data, and which surpasses its teacher, GPT-4, on STEM-focused question answering.
As of October 2026

Phi-4-reasoning-vision at AI Frontiers

In March 2026, they co-lead Phi-4-reasoning-vision-15B, a compact open-weight multimodal reasoning model for math and science reasoning and for understanding user interfaces, and coauthor its Microsoft Research blog post with John Langford, Neel Joshi and others. They are a Member of Technical Staff on the Microsoft Research AI Frontiers team.