Noriyuki Kojima

I am a Co-Founder and CEO of Kotoba Technologies, Inc.

Previously, I graduated from CS Ph.D at Cornell University in 2023, where I was advised by Yoav Artzi. I am grateful that My Ph.D. study was supported by Cornell Hopcroft Fellowship (2019-2020) and Masason Fellowship (2020~2022). I was a Research Scientist Intern at Hugging Face in summer 2022, working with Sasha Rush. and Victor Sanh; I was a Research Specialist at Princeton University in 2019, working with Jia Deng; I did my undergraduate at the University of Michigan from 2015 to 2018 with B.S.E in Computer Science where I worked with Jia Deng, Rada Mihalcea and Dragomir Radev; I was a Software Engineer Intern in the Facebook Applied Machine Learning team in summer 2017, where I was supervised by Juan Pino.

Email  /  Google Scholar  /  LinkedIn  /  Twitter

I'm interested in natural language processing, computer vision, and machine learning. One of the core topics in my research is about how to learn grounded language systems thorugh interactions with human users. Representative papers are highlighted.

Presenting a framework to jointly study vision and language tasks and the fundamental steps of phrase grounding.

Presenting KILOGRAM, a resource for studying abstract visual reasoning in humans and machines.

Presenting a new benchmark for language-conditioned reinforcement learning in visual environments.

Studying diffusion models for markup-to-image generation.

Studying continual learning for natural language instruction generation by observing human users' instruction execution in a collaborative game.

Exploring what is learned by the unsupervised vision and language constituency parser via extreme simplification of model architectures.

Presenting Open Annotations of Single Image Surfaces (OASIS): a dataset for images in the wild with dense annotations of detailed 3D geometry.

Representing character names by adding social networks of discourse in embedding objectives.

Comparing learning-based methods and classical methods for navigation in virtual environments. Classical methods > learning-based methods in high-level, but learning-based methods learned useful regularities.

Proposing a new model for speaker naming in movies that leverages visual, textual, and acoustic modalities in an unified optimization framework.

Proposing structured matching of phrases and regions that encourages the semantic relations between phrases to agree with the visual relations.

Last updated: August 2021
The template was borrowed from Jon Barron's implementaion.