Active Transfer Learning (ATL) | Apply Online

Active Transfer Learning (ATL) - ONR Grant Opportunity FAQs

1) What is the Active Transfer Learning (ATL) opportunity?

Active Transfer Learning (ATL) is an Office of Naval Research (ONR) grant opportunity supporting basic research that advances autonomy and automation in dynamic, unconstrained real-world environments. The emphasis is on learning systems that remain robust when conditions change, data is imperfect or limited, and constant human supervision is not available.

2) What is the main research problem ATL is trying to solve?

ATL targets the gap between what learning systems know from prior tasks and what they need to succeed in a new, changing operational setting. The program focuses on systems that can reuse prior knowledge effectively, detect when that knowledge is insufficient, and then make smart, limited use of subject matter expert (SME) input to adapt while operating.

3) How does ATL relate to transfer learning and active learning?

ATL is positioned as a deliberate fusion and extension of transfer learning and active learning. Transfer learning is about reusing knowledge from earlier tasks/domains to improve learning in a new one. Active learning is about learning efficiently by strategically asking for help (often labels) on selected examples. ATL aims to combine both while addressing their practical limitations in real deployments.

4) What limitations of traditional active learning does ONR highlight?

The solicitation notes that traditional active learning often reduces the expert to a labeler. ATL seeks approaches that capture richer SME contributions, such as rules of thumb, causal explanations, operational constraints, and domain structure, rather than only requesting labels.

5) What limitations of traditional transfer learning does ONR highlight?

ONR notes that transfer learning can fail in practice, and when it fails there is often no practical, in-the-moment mechanism for an expert to diagnose the failure, repair the transfer, or fill in missing knowledge while the system is operating. ATL is intended to make transfer more diagnosable and correctable in situ with SME assistance.

6) What are the two main technical goals of the ATL program?

The solicitation describes two primary goals:

• Robust transfer using occasional SME availability: Methods that leverage an SME when available to improve transfer from existing sources, including machine-initiated and human-guided exploration of relevant prior knowledge.
• Injecting new knowledge during operation: Methods that allow targeted queries and enable SMEs to provide rich knowledge directly (not just labels) while the system is operating in the target setting.

7) What kinds of "prior knowledge" does ATL expect systems to use?

ATL explicitly points to knowledge that a machine can access from past tasks, existing models, repositories, datasets, or other sources. The program calls out capabilities such as searching, selecting, and adapting models, data, or structured knowledge as part of machine-initiated exploration.

8) What does ONR mean by "occasional" access to a subject matter expert (SME)?

ATL assumes the SME will not be continuously available. Instead, expertise may be available intermittently and in-the-moment. The learning system is expected to decide when it is worth interrupting or querying the SME, rather than asking indiscriminately.

9) What types of SME input are encouraged beyond labeling?

The opportunity emphasizes higher-bandwidth forms of SME knowledge, including explanations, guidance, rules of thumb, causal insights, operational constraints, domain structure, and incremental refinement, similar to how a human apprentice might be taught.

10) What is meant by "in situ" diagnosis and correction?

In this context, "in situ" means while the system is operating in the target environment. The program seeks mechanisms for identifying when transfer is failing or incomplete and then correcting or completing it during operation with SME assistance.

11) Does ATL require the system to reason about what it knows and what it does not know?

Yes. A key component described is the machine's ability to reason about whether its current knowledge is sufficient and to make decisions about when to query the SME.

12) What kinds of outcomes is ATL ultimately aiming for?

ATL is research-focused and aims to produce learning capabilities that are more practical and resilient than standard transfer learning or active learning alone, especially in operational contexts where autonomy matters and conditions can shift unexpectedly.

13) What is the BAA number for this ONR opportunity?

The opportunity was issued as ONR BAA number ONRBAA10-017.

14) What is the CFDA number and program category listed?

The opportunity lists CFDA 12.300 (Basic and Applied Scientific Research) and is under the science and technology research and development category.

15) What funding instruments are listed for ATL?

Funding instruments listed include cooperative agreements, grants, and potentially other mechanisms such as procurement contracts. This suggests ONR may tailor the award mechanism based on the nature of the work and the level of collaboration or oversight required.

16) Is cost sharing or matching required?

No cost sharing or matching requirement is stated in the provided opportunity information.

17) What were the key dates and deadlines?

• Posting date: April 22, 2010
• White paper deadline: June 28, 2010 at 2:00 PM Eastern Time
• Full proposal deadline: September 4, 2010 at 2:00 PM Eastern Time
• Archive date: October 4, 2010

18) Who is eligible to apply as a prime applicant?

Eligibility is broadly open to responsible sources in academia and industry. Historically Black Colleges and Universities (HBCUs) and Minority Institutions (MIs) are explicitly encouraged to participate, although no funds are set aside specifically for them.

19) Are FFRDCs eligible to receive awards directly?

No. Federally Funded Research and Development Centers (FFRDCs), including Department of Energy national laboratories, are not eligible to receive awards directly under this opportunity.

20) Can Navy labs, warfare centers, or other federal labs submit as prime applicants?

No. Navy labs, warfare centers, and other federal labs cannot submit as prime applicants according to the provided information.

21) Can ineligible federal organizations participate at all?

Yes. FFRDCs and federal labs may participate through teaming arrangements with eligible proposers, subject to the rules of their sponsoring agreements.

22) Are there any export control or citizenship-related participation limits?

Some topics may involve export-controlled technologies. In those cases, participation may be limited to U.S. persons as defined by ITAR (22 CFR 120.1 et seq.).

23) Does this opportunity emphasize basic research, applied research, or deployment?

The opportunity is described as supporting basic research, with the goal of enabling more robust and practical learning approaches for autonomy in real-world environments.

24) What is the role of machine-initiated versus human-guided exploration in ATL?

The program calls for both: machine-initiated exploration (such as searching for and adapting relevant prior models/data/knowledge) and human-guided exploration where the SME can direct the system to the right sources or help interpret what is missing.

25) What does it mean to "inject new knowledge" in the target setting?

It means enabling the system to acquire and incorporate new domain knowledge while operating in the target environment, including through targeted questions and mechanisms for SMEs to supply rich, structured guidance beyond simple labels.