Research Resources
ARNI Suite: NeuroAI Models, Benchmarks, and Research Infrastructure
| Function | Tool | Description | Link |
| Neurofoundation Models & Infrastructure Azabou, (CU) | torch_brain | Framework for building and fine-tuning large-scale neurofoundation models using transformers. | https://github.com/neuro-galaxy/torch_brain |
| Neurofoundation Models & Infrastructure Azabou, (CU) | POYO+ | A multi-task, multi-session transformer model for neural decoding, trained on large-scale neural data to enable generalization across brain regions, cell types, and tasks, with ARNI-supported contributions to its development. | https://poyo-plus.github.io |
|
Neurofoundation Models & Infrastructure, Tolias (Stanford) New in Y3 |
OmniMouse | A large-scale, multi-modal neurofoundation model trained on over 150 billion neural tokens, enabling unified neural prediction, behavioral decoding, and forecasting across diverse tasks, and revealing distinct scaling laws for brain data | https://github.com/enigma-brain/omnimouse |
| Neurobehavioral Modeling & Representation Learning
Paninski (CU) New in Y3 |
BEAST (Behavioral Analysis via Self-Supervised Transformers) | A self-supervised transformer-based framework for modeling animal behavior from video, enabling neural encoding, pose estimation, and action segmentation from unlabeled data. | https://github.com/paninski-lab/beast |
|
Datasets and Data Infrastructure (Neural, Clinical, Multimodal) Azabou, (CU) |
brainsets | Standardized collection of pre-processed neural datasets ready for model training and benchmarking. | https://github.com/neuro-galaxy/brainsets |
| Datasets and Data Infrastructure (Neural, Clinical, Multimodal)
Azabou, (CU) |
temporaldata | Data structures and utilities for managing multi-modal, multi-resolution time series data (e.g., neural, video, behavioral). | https://github.com/neuro-galaxy/temporaldata |
|
Datasets and Data Infrastructure (Neural, Clinical, Multimodal) Hirschberg (CU), Bojic (Nanyang Technological University) New in Y3 |
SMARTMiner & SMARTSpan Dataset | Framework and dataset for extracting clinical treatment plans from unstructured text | https://github.com/IvaBojic/SMARTMiner |
| Datasets and Data Management
Wu, Hirschberg, (CU) New in Y3 |
Akan Cinematic Emotions (AkaCE) Dataset | First multimodal emotion dialogue dataset for an African language | https://github.com/zehuiwu/Akan-Cinematic-Emotion |
|
Benchmarks and Evaluation Frameworks Ziweig, Hirschberg (CU) New in Y3 |
Mental Health Resources and Benchmark Website | Repository and evaluation framework for mental health datasets that defines quantitative and qualitative criteria for fairness, generalizability, and privacy compliance in clinical AI applications. | https://ziweig.github.io/mental-health-datasets-resources-review/ |
| Benchmarks and Evaluation Frameworks
Chen, Hirschberg (CU) New in Y3 |
SPEECHMENTALMANIP Benchmark | Synthetic multi-speaker benchmark for detecting manipulation in dialogue | https://github.com/runjchen/speech_mentalmanip |
|
Benchmarks and Evaluation Frameworks Zemel (CU) , McKeown(CU), Tolias (Stanford) New in Y3 |
SAUCE / Few-shot Evaluation Framework | A novel evaluation framework for continual learning that introduces few-shot metrics and the Scaled Area Under the Adaptation Curve (SAUCE) to quantify plasticity and rapid adaptation in vision and language models; currently under development as part of ARNI’s continual learning efforts. | in development |
| Benchmarks and Evaluation Frameworks
Zemel, McKeown (CU), Tolias (Stanford) New in Y3 |
“Day in the Life” Continual Learning Benchmark Suite | An emerging, cognitively inspired benchmark that models real-world task repetition and temporal structure, addressing limitations of standard continual learning benchmarks and enabling more realistic evaluation of adaptive AI systems; under active development within ARNI working groups. | in development |
|
Benchmarks and Evaluation Frameworks (McKeown) New in Y3 |
Day in Life Benchmark Suite-
LiveNewsBench |
A continuously updated benchmark for evaluating agentic web search in large language models, using recent news to assess multi-step retrieval, reasoning, and real-time information access beyond training data.It contributes to ARNI’s “Day in the Life” benchmark by capturing dynamic, real-world information-seeking tasks. | https://github.com/LiveNewsBench/LiveNewsBench/tree/main |
| Benchmarks and Evaluation Frameworks
(McKeown) New in Y3 |
Day in the Life Benchmark Suite- Multilingual Affective State Identification | A multilingual benchmark for evaluating how AI systems interpret affective states across eight languages, capturing culturally grounded expressions of emotion to assess cross-linguistic understanding and continual learning. | in development |
|
Benchmarks and Evaluation Frameworks (Zemel,Miller, Richards, Pitkow) New in Y3 |
Hyper-Modal Representation Learning Benchmark | A benchmark for learning multimodal representations without supervision, evaluating biologically inspired algorithms across three tracks: multimodal category discovery, large-scale representation learning, and agentic reasoning. | in development |
|
Representation Learning Shan (CU) New in Y3 |
Connectome Embedding | A framework for learning low-dimensional representations of brain connectivity data, enabling analysis of neural structure and function and supporting biologically informed machine learning models. | https://github.com/hzshan/connectome_embedding |
| Embodied AI & Brain–Behavior Modeling
Richards (MILA) Olveczky (Harvard) New in Y3 |
MIMIC-MJX | An open-source framework for physics-based simulation of animal behavior, integrating neural, behavioral, and biomechanical data, with datasets, pretrained models, and tools for motion tracking and analysis. | https://mimic-mjx.talmolab.org/ |
|
Embodied AI & Brain–Behavior Modeling Chaudhari, Balasubramanian (UPenn) New in Y3 |
REMI (Reconstructing Episodic Memory in Navigation) | Biologically grounded framework linking hippocampal–entorhinal (HC–MEC) circuits to memory-driven planning and navigation, implemented in simulated environments (RatatouGym, Habitat) for evaluating spatial reasoning and adaptive behavior | https://zhaozewang.github.io/remi |
|
Human-AI Interaction & Cognitive Modeling Tools Toosi (CU) New in Y3 |
HuggingFace Demo – Human Hallucination Prediction | Predicts human perceptual hallucinations from visual input | https://huggingface.co/spaces/ttoosi/Human_Hallucination_Prediction |
|
Human-AI Interaction & Cognitive Modeling Tools (Toosi (CU) New in Y3 |
HuggingFace Demo – Generative Inference (Perceptual Organization) | Models what humans perceive based on perceptual organization laws | https://huggingface.co/spaces/ttoosi/GenerativeInferenceDemo |
| Human-AI Interaction & Cognitive Modeling Tools
Shi (UPenn) New in Y3 |
VibeSpace | An interactive visualization tool for exploring learned representations, enabling analysis of structure and semantic relationships in high-dimensional embedding spaces. | https://huggingface.co/spaces/huzey/VibeSpace |
|
Embodied AI & Physical Systems Zolfaghari (U Memphis), Ebrahimi (VSU), Pitkow (CMU) New in Y3 |
Electromagnetic Soft Actuator Modeling Code | Analytical modeling and control of electromagnetic soft actuators | https://github.com/NafisEbrahimi/Analytical-Modeling-for-ESA |
| Model Training / Representation Learning
Noor (Tuskegee) New in Y3 |
SSL-SAR-ATR (Self-Supervised Learning for SAR Target Recognition) | A self-supervised learning framework for representation learning in synthetic aperture radar (SAR) imagery, enabling robust target recognition and generalization under limited labeled data. | https://github.com/MdAlSiam/ssl-sar-atr-2-v2/ |
|
Optimization and Model Analysis (Shi, UPenn) New in Y3 |
ncut-pytorch | A PyTorch implementation of normalized cuts for scalable clustering and segmentation, enabling efficient detection of structure in high-dimensional data and learned representations. | https://ncut-pytorch.readthedocs.io/ |
|
Optimization and Model Analysis (Shi, UPenn) |
ncut-pytorch | PyTorch implementation of normalized cuts to detect modular structure in learned neural representations. | https://ncut-pytorch.readthedocs.io/ |
| Experimental Tools (Issa - CU) | MkTurk | Web-based platform for running neuroscience and behavioral experiments online. | https://github.com/issalab/mkturk |
|
Electrophysiology Tools (Issa - CU) |
DREDge | Tool for robust motion correction in high-density extracellular recordings across different species. | https://github.com/evarol/dredge |
| Model Training (McKeown, CU) |
SPiCy: Unsupervised sparse predictive coding | New metric for evaluating detailed image captions generated by VLMs, combining scene graphs and LLMs-as-a-Judge to evaluate caption quality. | https://anonymous.4open.science/r/spicy-56D4 |
|
Community Engagement (Azabou - CU) |
COSYNE 2025 Tutorial: Transformers in Neuroscience | A tutorial focused on the application of transformer models in neuroscience. | https://cosyne-tutorial-2025.github.io |
| Model Training (Chaudhari, UPenn) |
Prospective Learning: Principled Extrapolation to the Future | Framework for extrapolating future states in neural networks for prospective learning. | https://github.com/neurodata/prolearn |
|
Tutorials/Training (Chaudhari, UPenn) |
ProLearn Tutorials | Tutorial for implementing prospective learning techniques in neural networks. | https://github.com/neurodata/prolearn/blob/main/tutorials/tutorial.ipynb |
| Representation Learning
(Chaudhari, UPenn) |
Time Makes Space: Place Fields from Episodic RNNs | Explores the emergence of place fields in networks encoding temporally continuous sensory experiences. | https://github.com/zhaozewang/place_cells_episodic_rnn |
Get ACCESS Through ARNI
Do you need better computing resources for your project, research, or interest? ARNI is offering subscriptions to the NSF The ACCESS Allocations Service.
Get NAIRR Pilot
"The NAIRR Pilot aims to connect U.S. researchers and educators to computational, data, and training resources needed to advance AI research and research that employs AI. Federal agencies are collaborating with government-supported and non-governmental partners to implement the Pilot as a preparatory step toward an eventual full NAIRR implementation."
