Project summary

I am an undergraduate student requesting funding to attend NeurIPS 2026 in Sydney, Australia, where I hope to share my research on one-shot learning and connect with researchers working on AI safety.

Our recent paper One-shot Familiarity Learning is Facilitated by Visual Category Structure Learned through Training Augmentation was accepted at the NeurIPS 2026 NeurReps workshop.

My work studies how neural networks form memories after a single exposure and how the structure of their internal representations affects recognition. I've found that training with image augmentations improves memory performance and produces representations with properties similar to neurons in the human medial temporal lobe.

I'm interested in extending this work to understand failure modes in AI systems with persistent memory. In particular, I want to study whether the representations that help models recognize previously seen information also make them vulnerable to false recognition or adversarial manipulation. NeurIPS would be a valuable opportunity to get feedback on this direction and find collaborators working on related problems.

What are this project's goals? How will you achieve them?

My main goal is to develop my existing work on one-shot familiarity learning into a research direction with concrete applications to AI safety.

At NeurIPS, I want to meet researchers working on memory-augmented AI agents, mechanistic interpretability, and adversarial robustness. I'm particularly interested in understanding how current agent memory systems handle conflicting or manipulated information, and whether insights from biologically inspired memory models can help explain their failure modes.

I also want feedback on my existing paper and ideas for follow-up experiments. One direction I'm considering is testing whether representations that generalize well across different views of the same input are also more likely to confuse similar but distinct inputs.

My goal is to leave the conference with a clearer experimental plan, connections with researchers working on these problems, and potential collaborators for a follow-up paper.

How will this funding be used?

$500 -- conference registration

$1000 -- travel funding to the conference (flight to Sydney, Australia from the US)

Who is on your team?

I have one co-first author fellow undergrad student on my team. Our background is in computational neuroscience and machine learning, with a focus on how neural networks learn and represent memories.

I hope to build on this experience by connecting work on memory representations with research on the safety and reliability of AI agents.

What are the most likely causes and outcomes if this project fails?

If we do not receive funding, we will be unable to visit and present our work at the conference, and to connect with fellow researchers and communicate our results.

How much money have you raised in the last 12 months, and from where?

None so far.