Amir Arsalan Soltani

I am a research assistant in Professor Josh Tenenbaum's Computational Cognitive Science Lab at Massachusetts Institute of Technology, where I work on building computational models for perceptions inspired by cognitive science to endow future AI agents with visual intelligence.

I graduated in 2016 with a Master's Degree in Computer Science from State University of New York at Buffalo. Prior to that, I did my undergraduate at Isalamic Azad University in Iran, where I received my B.S. in Computer Software Engineering.

Email  /  CV  /  Google Scholar  /  GitHub  /  LinkedIn  /  Twitter


My long-term research goal is to enable AI agents to have the ability to imagine (produce mental or abstract imagery without sensory inputs) by building models of the visual world and come up with concise and generalizable theories/solutions that can explain phenomena beyond what low-level statistics of observable data spells.


Perceiving Fully Occluded Objects via Physical Simulation (tentative title)
Yildirim, I.*, Siegel, M.*, Soltani, A.*, Chaudhuri, S. & Tenenbaum, J.
(* indicate equal contribution)
Manuscript in Preparation, 2019

A model-based, compositional perception system for recovering 3D shapes covered by cloth, with low sample complexity.

The Little Prince, Uncover the truth

Draping an Elephant: Uncovering Children's Reasoning About Cloth-Covered Objects
Ullman T., Kosoy E., Yildirim I., Soltani AA., Siegel M., Tenenbaum J. & Spelke E.
Cognitive Science Society (CogSci) , 2019
Paper (PDF)   

This work is a contribution towards The Core Object System. Here, we show that preschool children are able to use their "intuitive" physical knowledge of how cloths work and reason about what shape has been draped by cloth, hence uncovering the true object.

3D generation, 3D reconstruction
Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes with Deep Generative Networks
Soltani, AA., Huang, H., Wu, J., Kulkarni, T., & Tenenbaum, J.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2017
Paper (PDF)  /  Code  /  Poster  /  Slides (includes more results)

A generative model for generic 3D shapes to obtain abstract description of objects as a crucial component for building models of the environment through inverse graphics.

website template credits