ARNI completed its first year of operations, reaching significant milestones: 

Fifteen research projects were funded, representing a vibrant network of collaborations among all sites. 

• 6 projects funded are Columbia-based projects, 3 at UPenn, 2 at CUNY, 1 at Tuskegee, 1 at Yale, 1 at Carnegie Mellon University, and 1 at MILA
• 27 faculty, 12 graduate students, 4 postdoctoral fellows, and 1 research technician were funded to work on these projects
• These projects have already generated ten publications, with numerous manuscripts under preparation

NeuroAI Course: In collaboration with the Neuromatch Academy, a new field-defining course will be launched in July. 450 students from 75 countries and from all continents applied to participate!

ARNI also collaborates with the New York Hall of Science and student-led groups at Columbia to develop and deliver hands-on activities on neuroscience and AI concepts! 

In our second year, we look forward to working with all of you to broaden our activities and strengthen the ARNI community connections in more impactful ways!

"The Relational Bottleneck as an Inductive Bias for Efficient Abstraction."

By: Taylor W. Webb, Steven M. Frankland, Awni Altabaa, Simon Segert, Kamesh Krishnamurthy, Declan Campbell, Jacob Russin, Tyler Giallanza, Zack Dulberg, Randall O'Reilly, John Lafferty, and Jonathan D. Cohen

Abstract: A central challenge for cognitive science is to explain how abstract concepts are acquired from limited experience. This has often been framed in terms of a dichotomy between connectionist and symbolic cognitive models. Here, we highlight a recently emerging line of work that suggests a novel reconciliation of these approaches, by exploiting an inductive bias that we term the relational bottleneck. In that approach, neural networks are constrained via their architecture to focus on relations between perceptual inputs, rather than the attributes of individual inputs. We review a family of models that employ this approach to induce abstractions in a data-efficient manner, emphasizing their potential as candidate models for the acquisition of abstract concepts in the human mind and brain.

"Brain Decodes Deep Nets"

By: Huzheng Yang, James Gee, Jianbo Shi

Abstract: We developed a tool for visualizing and analyzing large pre-trained vision models by mapping them onto the brain, thus exposing their hidden inside. Our innovation arises from a surprising usage of brain encoding: predicting brain fMRI measurements in response to images. We report two findings. First, explicit mapping between the brain and deep-network features across dimensions of space, layers, scales, and channels is crucial. This mapping method, FactorTopy, is plug-and-play for any deep-network; with it, one can paint a picture of the network onto the brain (literally!). Second, our visualization shows how different training methods matter: they lead to remarkable differences in hierarchical organization and scaling behavior, growing with more data or network capacity. It also provides insight into fine-tuning: how pre-trained models change when adapting to small datasets. We found brain-like hierarchically organized network suffer less from catastrophic forgetting after fine-tuned.

ARNI YEAR 1 Publications

Altabaa, A., & Lafferty, J. (2024a). Approximation of relation functions and attention mechanisms (arXiv:2402.08856). arXiv. http://arxiv.org/abs/2402.08856
https://doi.org/10.48550/arXiv.2402.08856

Altabaa, A., & Lafferty, J. (2024b). Learning Hierarchical Relational Representations through Relational Convolutions (arXiv:2310.03240). arXiv. http://arxiv.org/abs/2310.03240
https://doi.org/10.48550/arXiv.2310.03240

Altabaa, A., Webb, T., Cohen, J., & Lafferty, J. (2023). Abstractors and relational cross-attention: An inductive bias for explicit relational reasoning in Transformers (arXiv:2304.00195). arXiv. http://arxiv.org/abs/2304.00195
https://doi.org/10.48550/arXiv.2304.00195

Chiquier, M., Mall, U., & Vondrick, C. (2024). Evolving Interpretable Visual Classifiers with Large Language Models. (In submission to European Conference on Computer Vision (ECCV) 2024). https://doi.org/10.48550/ARXIV.2404.09941

Eyre, B., Creager, E., Madras, D., Papyan, V., & Zemel, R. (2023). Out of the Ordinary: Spectrally Adapting Regression for Covariate Shift (arXiv:2312.17463). arXiv. http://arxiv.org/abs/2312.17463
https://doi.org/10.48550/arXiv.2312.17463

H. Yang, J. Gee, and J. Shi. Brain decodes deep nets. CVPR, Spotlight, 2024. arXiv.2312.01280.
https://doi.org/10.48550/arXiv.2312.01280

Mahdaviyeh, Y., Lucas, J., Ren. M., Tolias, A., Zemel, R., Pitassi,. T. (2024). Replay Can Probably Increase Forgetting. Submitted to NeurIPS.

McGaughey, K. D., & Gold, J. (2023). Neuroscience 2023.

Contributions of sensory adaptation and pupil-linked arousal to perceptual decisions about uncertain and unstable visual stimuli. Society for Neuroscience.

Tyulina, N., Emmanouil, T. A., & Levitan, S. I. 2024. ACM Conversational User Interfaces 2024. In Understanding Linguistic and Visual Factors that Affect Human Trust Perception of Virtual Agents. Luxembourg City.

Webb, T. W., Frankland, S. M., Altabaa, A., Segert, S., Krishnamurthy, K., Campbell, D., Russin, J., Giallanza, T., O’Reilly, R., Lafferty, J., & Cohen, J. D. (2024). The Relational Bottleneck as an Inductive Bias for Efficient Abstraction. Accepted in Trends in Cognitive Science, 2024. (arXiv:2309.06629). arXiv. http://arxiv.org/abs/2309.06629)
https://doi.org/10.48550/arXiv.2309.06629

ARNI in the News!

"Exploring the Frontier of AI and Quantum Computing"

ARNI's "Breakthrough Technologies" panel was featured in Columbia Engineering News. Please find the full article by using this link:
https://www.engineering.columbia.edu/news/exploring-frontier-ai-and-quantum-computing

ARNI's Annual Retreat

Please stay tuned for our annual retreat.  It will be a two day event featuring the past year's publications, research, and successes. We will also be introducing the new projects, educational endeavors, and exciting future events to come.

Dates: October 21 & 22
Location:TBD
Times: 9am to 5pm 

You can find any upcoming seminars and events on ARNI's website, https://arni-institute.org/events/. 

NeuroAI Activities

ARNI, partnering with Columbia University Neuroscience Outreach, Columbia STEM Starters, and NYSCI, is developing educational activities focused on neuroAI. Tailored for ages K-12, these activities aim to stimulate young minds into thinking about the operations of neurons and the decision-making process of the brain. Mastering these fundamental concepts are the initial steps toward understanding basic components of how AI technology works.

Please fill out ARNI’s demographic survey: 
https://seas.co1.qualtrics.com/jfe/form/SV_389MBAs7tPell5Q

ARNI Working Groups
(all events are on the ARNI events page)

Animal Behavior Video Analysis Working Group: (monthly - student and faculty-led) focuses on understanding the neural control of movement, cognition, and social interaction, and the tools currently used to precisely quantify motor behaviors. The group examines the limitations of standard approaches and incorporates presentations of arni faculty. Srini Turaga (Janelia), Jianbo Shi (Upenn), and Ugne Klibaite (Harvard) were invited to present their work in person at CU.

Continual Learning and Frontier Models Working Group: (weekly-student-led) focuses on continual learning in large language and vision models. The group reviews tutorials on the current continual learning setting and approaches, followed by most recent research mostly on LLMs. Students from CU (Zemel, McKeown, Bareinboim) and Princeton (Griffiths group) help lead this group. They also invite speakers within CU and other Universities like NYU, to give a talk on their research. 

Multi-resource-cost Optimization of Neural Network Models Woking Group: (monthly) focuses on understanding biological neural mechanisms that emerge from particular profiles of resource costs and behavioral affordances and also engineering more efficient AI for resource-limited devices. It is led by the group of Niko Kriegeskorte at CU, while faculty and trainees from CU, Harvard, Stanford, CMU, Janelia participate and present.

ARNI Emerging Researchers
We have started! Get ready for some innovative and creative activities to come from the ARNI graduate students and postdoctorate fellows.

Benjamin Eyre (Richard Zemel group at Columbia), Katherine Xu (Jianbo Shi group at UPenn), and Dongrui Deng (Xaq Pitkow group at Carnegie Mellon) have graciously volunteered to spearhead this group. 

Please complete this interest form, only if you are a ARNI trainee and/or student. 

"Dear Scientist,

The annual Eppendorf & Science Prize for Neurobiology is an international prize which honors young scientists for outstanding neurobiological research based on methods of molecular, cellular, systems, or organismic biology. Researchers who are not older than 35 years are invited to apply.

Are you or one of your colleagues doing great neuroscience? If so, then we encourage you to apply for the prestigious Eppendorf & Science Prize for Neurobiology.

Or do you know someone who might be eligible to apply? If so, please help us spread the news!

Read below for what’s in store for the Grand Prize Winner! Good to know - up to three finalists are honored, too!"

Your Eppendorf & Science Prize Team

 The Grand Prize Winner receives 

• Prize money of US$25,000
• Publication in Science of an essay by the winner about his/her research
• Full support to attend the Prize Ceremony held in conjunction with the Annual Meeting of the Society for Neuroscience in the USA
• 10-year AAAS membership and online subscription to Science
• Complimentary products worth US$1,000 from Eppendorf
• An invitation to visit Eppendorf in Hamburg, Germany

Deadline: June 15, 2024

We would like to highlight ARNI achievements in future newsletters. Please share with us your events, papers, presentation, or any news you want to share with the ARNI community!