2026 KAI-X Neuromorphic Computing Workshop

• When: Nov 11th Wednesday, 1pm-5pm

• Where: KAIST main campus, W13 building auditorium (Meta Convergence Bldg. I, 서측 메타융합관)

1:00pm Opening

1:10-1:50pm
Brain-Inspired Hardware for Next-Generation Computing
(Prof. Dongsuk Jeon, Seoul National University)

Neuromorphic computing has attracted serious interest across various research fields ranging from computational neuroscience to artificial intelligence. However, this brain-like approach also offers a great opportunity for designing efficient hardware, not only due to how it computes, but also due to its fundamentally different architecture. This talk discusses the opportunities neuromorphic computing can bring to next-generation computing platforms. It also covers the critical challenges hindering practical applications of neuromorphic computing, as well as ongoing efforts to overcome those hurdles.

1:50-2:30pm
From Neural Circuits to Intelligent Machines
(Prof. Se-Bum Paik, KAIST)

Biological brains exhibit remarkable robustness and adaptability that remain challenging to achieve in artificial neural networks. This talk explores how key principles discovered through brain research can inspire solutions to fundamental limitations of current deep neural networks. First, we show how prewired, static visual receptive fields, inspired by early visual processing in the brain, enable robust object recognition across changing environments and improve continual learning by promoting shape-based representations. Second, we introduce a neuromimetic metaplasticity model inspired by working memory, in which synapses with different degrees of flexibility enable continual learning while mitigating catastrophic forgetting. This model dynamically balances memory stability and flexibility without requiring additional training procedures or architectural modifications. It also enhances robustness to data poisoning by selectively retaining significant information. Together, these studies demonstrate how principles intrinsic to biological neural circuits can address key challenges in artificial intelligence, offering new directions for the development of more robust, adaptive, and brain-inspired intelligent machines.

2:30-3:10pm
Building a Memory Hierarchy for Analog Compute-in-Memory through Device–Algorithm Co-Optimization
(Prof. Sangbum Kim, Seoul National University)

We propose an ACiM memory hierarchy composed of two complementary synaptic-device technologies: one optimized for inference and the other for training. Efficiently integrating these devices requires close device–algorithm co-optimization. We explore three complementary strategies: (1) identifying existing algorithms that are inherently tolerant to device imperfections, (2) adapting conventional algorithms to better exploit the characteristics of analog hardware, and (3) developing new algorithmic frameworks that are intrinsically aligned with device physics. These studies demonstrate that a heterogeneous memory hierarchy can provide a practical pathway toward efficient and scalable ACiM systems. More broadly, they highlight that tightly coupling device design with algorithm development—rather than optimizing either in isolation—is essential for fully realizing the potential of analog memory-based AI computing.

3:10-3:50pm
Credit assignment as a forward-inverse problem, task generation as an inverse-forward solution
(Prof. Sang Wan Lee, KAIST)

The brain navigates dynamic environments by predicting future outcomes from present observations transmitting predictive signals forward across space and time. Discrepancies between prediction and observation guide a complementary, backward-propagating error signal that continuously updates the internal representation. This forward-inverse process underlies temporal credit assignment. Building on this framework, I recast task generation as an inverse-forward problem. Through proof-of-concept demonstrations, I will illustrate how this dual perspective challenges traditional paradigms of task design and modelling, offering a new way to study computational principles in neural circuits.

4:00-5:00pm *BCS colloquium talk
Credit assignment in physical neuronal networks
(Prof. Mihai A. Petrovici, University of Bern)

Whether biological or artificial, intelligence ultimately boils down to the ability of physical substrates to perform (complex) computations efficiently. Which algorithms can "run" on the dynamics of a given substrate? And conversely, which substrates should we build to implement our algorithms of choice? In my talk, I will discuss variants of these questions that are inspired by recent insights from neurobiology and recent developments in neuromorphic engineering. In particular, I will address several challenges of credit assignment in physical neuronal networks and show how they can be met with rigorous math and creative physics.

5:00-5:15pm Discussion

This event is sponsored by KAI-X and KAIST Center for Neuroscience-inspired AI.