Forest Agostinelli

Forest Agostinelli

Assistant Professor - AI Institute - Computer Science and Engineering
University of South Carolina
Email: foresta@cse.sc.edu

I am an assistant professor at the University of South Carolina. My research goal is to use artificial intelligence to solve planning problems, explain generalized solutions to humans, and collaborate with humans to automate the discovery of new knowledge. The methods employed in my research group include deep learning, reinforcement learning, heuristic search, and formal logic.

I earned my Ph.D. in computer science at the University of California, Irvine, my M.S. in computer science at the University of Michigan, and my B.S. in electrical and computer engineering at The Ohio State University.

Prospective Ph.D. students can send an email to foresta@cse.sc.edu with all of the following documents:
1) A short writeup on how your background fits in to one of the following papers and what future research directions excite you (1,2,3,4)
2) CV
3) Transcripts 

Updates

  • 09/2024: I will be presenting my research at the RLeap Symposium.
  • 09/2024: Awarded an NSF grant from Robust Intelligence on Scalable Heuristic Learning
  • 05/2024: One paper accepted at the Reinforcement Learning Conference 2024 
  • 4/2024: Four workshop papers accepted at ICAPS 2024: PRL (1,2), HAXP, HSDIP
  • 2/2024: I will be giving a master class on deep learning, reinforcement learning, and heuristic search at SoCS 2024.
  • 2/2024: One paper accepted to ICAPS 2024
  • 09/2023: Awarded an NSF-MRI grant 
  • 05/2023: Awarded an ASPIRE-I grant from the Office of the Vice President for Research
  • 04/2023: Awarded a MADE in SC grant from the South Carolina EPSCOR program
  • 11/2022: Two papers (1, 2) published in HCI International 2022 - Late Breaking Posters
  • 06/2022: One paper published in Nucleic Acids Research
  • 06/2022: One paper published in the AAAI 2022 - Demonstration Track
  • 02/2022: One paper published in Nature Communications
  • 04/2021: Awarded two ASPIRE-II grants from the Office of the Vice President for Research
  • 03/2021: One paper published in Neural Networks
  • All Updates

Current Projects

Forest Agostinelli Rubik's cube Research University of South Carolina

Specifying Goals to Deep Neural Networks

For practical use, when solving planning problems, deep neural networks must be able to generalize over goals without the need for retraining. To accomplish this, we incorporate formal logic with the training of deep neural networks and with heuristic search.
Selected Papers:
Specifying Goals for Deep Neural Networks with Answer Set Programming, ICAPS (2024)

A Conflict-Driven Approach for Reaching Goals Specified with Negation as Failure, ICAPS HAXP Workshop (2024)

Forest Agostinelli Rubik's cube Research University of South Carolina

Learning World Models for Planning

Many planning problems in real-world domains are not given world models and, thus, cannot make use of planning. We seek to learn world models from real-world observations and use these learned world models to learn heuristic functions and to plan.
Selected Papers:
Learning Discrete World Models for Heuristic Search, Reinforcement Learning Conference (2024)

Forest Agostinelli Rubik's cube Research UCI UC Irvine

Solving Planning Problems with Deep Reinforcement Learning and Search

We seek to create domain-independent machine learning methods that can learn domain-specific heuristic functions given only a description of a planning domain. We have created DeepCubeA, a deep reinforcement learning and search algorithm capable of solving planning problems such as the Rubik's cube and other combinatorial puzzles without human guidance. We are investigating how DeepCubeA can be extended to problems in mathematics and the natural sciences.
DeepCubeA Webserver
Selected Papers:
Solving the Rubik's Cube with Deep Reinforcement Learning and Search, Nature Machine Intelligence (2019)

Solving the Rubik's Cube with Approximate Policy Iteration, ICLR (2019)



Forest Agostinelli Deep Learning Research UCI UC Irvine

Learning Activation Functions

Artificial neural networks typically have a fixed, non-linear activation function at each neuron. We have designed a novel form of piecewise linear activation function that is learned though gradient descent. With this adaptive activation function, we are able to improve upon deep neural network architectures that use static activation functions.
Selected Papers:
SPLASH: Learnable Activation Functions for Improving Accuracy and Adversarial Robustness, Neural Networks, 2021

Learning Activation Functions to Improve Deep Neural Networks, ICLR Workshop, 2015

Forest Agostinelli Circadian Rhythms Research UCI UC Irvine

Circadian Rhythms

Circadian rhythms are found in virtually all forms of life. They play a fundamental role in functions ranging from metabolism to cognition. We have developed Circadiomics for accessing and mining circadian omic datasets and BIO_CYCLE for analyzing cicadian rhythms experiments using deep learning.
Circadiomics Webserver
BIO_CYCLE Webserver
Selected Papers:
CircadiOmics: Circadian Omic Web Portal, Nucleic Acids Research (2022)

What Time is It? Deep Learning Approaches for Circadian Rhythms, ISMB (2016)



Forest Agostinelli Neuroscience Research UCI UC Irvine

The Hippocampus and Nonspatial Memory

The hippocampus plays a key role in the memory of sequences of events, however, the role of the hippocampus in nonspatial tasks has yet to be understood. Using unsupervised deep learning techniques, we visualize hippocampal activity during a nonspatial sequential memory task. We discovered that hippocampal activity correlates strongly with the sequence presented in this nonspatial memory task.
Selected Papers:
Hippocampal Ensembles Represent Sequential Relationships Among Discrete Nonspatial Events, Nature Communications (2022)

Research Gallery

Mobirise web page creator - Try here