Projects

Within this large brain network are 100 trillion synapses and 100 billion neurons, consuming only 20 watts of energy. To scale to this level of computing power, the next generation of AI computing power needs to be highly efficient.

• Brain-Inspired Algorithms - Accelerate the next generation of AI research by exploring under-explored mathematical algorithms from “brain-inspired” massively scalable approaches.

• Shared Learning - Develop greater abstracting and reasoning capabilities, such as applying knowledge from one domain to another, perceiving, and learning.

• Better Performance - This model must validate for a potential 10x improvement in energy efficiency and data rate handling capability.

Challenge Inspired: AI Next Campaign (https://www.darpa.mil/work-with-us/ai-next-campaign )

Performers will be expected to provide, at a minimum, the following deliverables. Negotiated deliverables specific to the proposed effort, such as reports, experimental and simulated data sets, proposed architectures, protocols, software codes, publications, model data, metrics, validation data, and other associated documentation and results.

1. Report on novel computational theory and related initial brain-inspired algorithms.

2. Report on updated expectations for energy efficiency improvements, data handling capabilities of the proposed approach, and preliminary discussion of potential hardware implementation.

3. Report on initial training and test data sets (simulated or modeled), evaluation metrics, and initial analysis results.

Challenge Inspired: AIxCC (https://aicyberchallenge.com/ )

Music performers will be expected to provide negotiated deliverables specific to the proposed effort, such as reports, experimental and simulated data sets, proposed architectures, protocols, software codes, publications, model data, metrics, validation data, and other associated documentation and results.

1. Create a novel computational music theory and related initial brain-inspired algorithms.

2. Report on updated expectations for energy efficiency improvements, data handling capabilities of the proposed approach, and preliminary discussion of potential hardware implementation.

3. Play on initial training and test data sets (simulated or modeled), evaluation metrics, and initial analysis results.

Challenge Inspired: Mixstreams (https://mixstreams.voracity.us/ )

Note. AI-generated image by svstudioart on Freepik.

Data architects and engineers will be expected to provide negotiated deliverables of data center designs specific to the proposed effort, such as reports, experimental and simulated data sets, proposed architectures, protocols, software codes, publications, model data, metrics, validation data, and other associated documentation and results.

1. Create a computing lab for generative AI experiments that will meet the current models with characteristics of elasticity of GPU hardware, software security, AI privacy, and micro-services (e.g., VMware ESXi).

2. Report on updated expectations for energy efficiency improvements, data handling capabilities of the proposed approach, and preliminary discussion of potential hardware implementation.

3. Play on initial training and test data sets (simulated or modeled), evaluation metrics, and initial analysis results.

Challenge Inspired: Meditron -70B (https://doi.org/10.48550/arXiv.2311.16079 )

Note. AI-generated image by Macro Eye Iris on Freepik.

AI Solutions Architects will be expected to perform image preprocessing, object detection, and data collection and analysis tasks. The provided negotiated deliverables of vision algorithm design specific to the proposed effort, such as reports, experimental and simulated data sets, proposed architectures, protocols, software codes, publications, model data, metrics, validation data, and other associated documentation and results.

1. Create a detection application for generative AI experiments that can detect preprocess test images. Research current models with characteristics of deployment onto GPU hardware (e.g., Nvidia), software (e.g., Nvidia CUDA), AI algorithm development software (e.g., Mathworks), and deployment edge AI devices (e.g., Nvidia Jetson, AGX Orin).

2. Report on updated expectations for energy efficiency improvements, data handling capabilities of the proposed approach, and preliminary discussion of potential hardware implementation.

3. Play on initial training and test data sets (simulated or modeled), evaluation metrics, and initial analysis results.

Challenge Inspired: Mathworks (https://www.mathworks.com/campaigns/offers/generate-cuda-gpu-code-matlab.html )

Note. Docker Desktop running gen AI locally in a contained environment.

AI Developers will be expected to perform various tests on several generative models to find gaps in logic (i.e., hallucinations). These tasks include developing a development platform to meet the needs by building generative AI models in a virtual environment. These virtual environments provided isolation and versioning flexibility, such as different versions of Python, Node.js, and Bash, to build a platform of services. The negotiated deliverables are to construct a web platform specific to the proposed effort to run generative AI locally. An experimental Web UI design, selection of gen AI datasets, proposed architectures, virtual environment software, publications, and other associated documentation are to be included in the report.

1. Create a generative AI web platform that can be run locally on-premise. Compare current models' deployment characteristics onto GPU hardware, AI algorithm development software, and deployment edge AI devices.

2. Report on updated expectations for energy efficiency improvements, data handling capabilities of the proposed approach, and preliminary discussion of potential hardware implementation.

3. Play on initial training and test data sets (simulated or modeled), evaluation metrics, and initial analysis results.

Challenge Inspired: Open WebUI (https://www.openwebui.com/ )