SBIR | Kalyan Perumalla

Artificial Intelligence Tools for Catalyzing Interdisciplinary Science (SBIR)

Sat, 01 Jul 2023 00:00:00 +0000

Solicitation PDF

Original at OSTI
Cached local copy

Selected Pages

Selected Extracts

This topic is focused on new artificial intelligence (AI) tools that enhance the productivity of scientists and engineers when making use of scholarly publications and engaging in interdisciplinary interactions in the areas of science and engineering supported by SC. While scientists are deeply knowledgeable in their areas of expertise, they are often not as highly informed in other important disciplines. For example, experts in several subareas of high-performance computing are not as deeply knowledgeable about the terminology, concepts, and state of the art in other areas such as biological sciences or high energy physics. The advent of new service-oriented access to technology that is based on large language models (LLM), which may include multi- modal data, has now opened the possibility of utilizing LLM-based commercial services to enable new interdisciplinary synergy. These services can now be envisioned to be used to digest large amounts of scientific publications and documentation across disciplines and enable interdisciplinary interactions that were not conceivable before.

Against this backdrop, grant applications are sought on the topic of “AI Tools for Catalyzing Interdisciplinary Science.” This topic will be focused on increasing the synergy among disciplines supported by the Office of Science. For example, this would include AI-based tools that can catalyze the interactions among scientists in nuclear physics and material sciences. Innovative methods are needed to assimilate the scientific publication corpus of two or more disciplines and enable scientists in any of those disciplines to collate scientific ideas, concepts, questions, and solutions from the other disciplines.

Included in scope is the integration with commercial AI services (Google, Microsoft, OpenAI, etc.) and open/commercial sources of publications and other data. Proposed approaches must display a short-term path to success and commercial viability. The proposed work should include a plan to perform demonstration activities with scientists and demonstrate verification and validation of results, including data validation.

Grant applications focused on the following will be considered out of scope:

Tools that address less than two scientific disciplines in Office of Science research areas.
Tools that build LLM-based services from scratch.
Security and hardening of LLM.

a. Interdisciplinary Training and Interfaces

Applications responsive to this subtopic will address the challenge of ingesting a multi-modal corpus of scientific publications of two or more scientific disciplines in Office of Science research areas into a knowledgebase to be built using LLM-based services.

Additionally, applications may address the creation of modern interfaces with natural language-based prompt- and-response support necessary for scientists to interact with the LLM back-ends of interdisciplinary knowledgebases.

Digital-Twin Capabilities for Science Network Infrastructures (SBIR)

Sat, 01 Jul 2023 00:00:00 +0000

Solicitation PDF

Original at OSTI
Cached local copy

Selected Pages

Selected Extracts

Digital twins are an emerging area of modern science where a physical object (i.e., device, process, or infrastructure) is paired with a digital (virtual) version of that same object. Operations of the physical object can generate data to validate the virtual objects behavior while the virtual object allows rapid exploration of input parameters that might damage the physical object. It is this close interaction between the physical and virtual object that makes this digital-twin environment so productive.

DOE has a long history of building and operating high performance physical network infrastructures. The Energy Science Network (ESnet) supports the Office of Science lab complex and it also peers with other research and education networks (RENs) both domestically and internationally. ESnet also operates an internal 100G SDN network testbed and the NSF funded FABRIC external network testbed. Specifically:

a) The ESnet 100G Software-Defined Networking (SDN) testbed’s objective is to provide network researchers with a realistic environment for testing. The current testbed enables 100G application / middleware experiments in addition to Science DMZ and SDN control/data plane experiments.

b) FABRIC: The National Science Foundation (NSF) collaboration is building a national research infrastructure that will enable the computer science and networking community to develop and test novel architectures that could yield a faster, more secure Internet. What is missing is a virtual companion, a digital twin, to these testbeds.

This topic solicits applications that would create the network simulation capabilities that would accurately and reliably duplicate the operational and performance capabilities of these testbeds creating their digital twin.

Mixed Integer Solver Technology for Accelerated Computing Systems (SBIR)

Sat, 01 Jul 2023 00:00:00 +0000

Solicitation PDF

Original at OSTI
Cached local copy

Selected Pages

Selected Extracts

This topic is focused on specific technologies that are required to advance the state of accelerated computing applied to mixed integer programming (MIP) problems arising in scientific applications. The primary focus of this topic is on the computer science needed to address the challenges in manifestation of scalable, distributed (multi-node) algorithmic techniques and not on combinatorial theory development.

MIP problems underlie many important application areas of interest to DOE, including biological systems, transportation networks, electric grids, and user facility infrastructures. While significant advances have been made in the theory and implementation of MIP solver methods on conventional central processing unit (CPU)- based hardware, new advances are necessary to fully utilize the DOE investments in accelerated computing platforms such as graphical processing unit (GPU)-based computers and high-performance computing systems. Preference may be given to applications that leverage existing ASCR software investments.

Grant applications focused on the following will be considered out of scope:

Cut generation, cut storage, cut manipulation methods,
Column generation methods,
Theory,
Fragments of technology that are isolated and cannot be demonstrated as part of a working MIP solver on standard problems such as found in the MIPLIB series, and
Solutions that cannot be demonstrated to run on GPU-based accelerators used in current or planned supercomputing systems of DOE leadership computing facilities.

Grant applications are sought in the following subtopics:

a. Efficient Distributed Tree Management This topic is focused on tree management that arises in branch-and-bound or branch-and-cut (B&C) methods to MIP solution methods. Research must focus on efficient representation and encoding of the B&C tree on accelerated memory hierarchies and address the challenges of efficient tree node movement, including exploitation of direct memory access (DMA) of accelerated memory across high-speed networks. Methods must solve the problem of scalable and efficient manipulation of B&C tree nodes. This includes the ability to query the quality of linear program relaxation, node ancestor identification, node deletion, and updates to the node data.

b. Efficient Linear Program Relaxation Solution Implementations of interior or exterior point methods must be developed specifically optimized for accelerated hardware using single-instruction-multiple-thread (SIMT) control flows or reconfigurable field programmable gate arrays (FPGAs). Methods must build on existing or new sparse and dense solvers and capable of static or dynamic (on-the-fly) choice of sparse versus dense solver based on the density of matrices20 Return to Table of Contents encountered in the input scenarios. Applicants may propose solving the linear program relaxations of MIP problem matrices that either fit entirely within a single node’s memory or large problem sizes where matrices do not fit within a single node’s memory but span multiple node memories. Milestones must aim to solve the relaxations of root nodes in increasing fractions (10%, 50%, 75%, 90%) of the problems in the MIPLIB 2017 problem set (with priming or probing methods not necessarily applied to the root problems).