Tag Archives: hippocampus

Striking similarity between memory processing of artificial intelligence (AI) models and hippocampus of the human brain

Leave a reply

A December 18, 2023 news item on ScienceDaily shifted my focus from hardware to software when considering memory in brainlike (neuromorphic) computing,

An interdisciplinary team consisting of researchers from the Center for Cognition and Sociality and the Data Science Group within the Institute for Basic Science (IBS) [Korea] revealed a striking similarity between the memory processing of artificial intelligence (AI) models and the hippocampus of the human brain. This new finding provides a novel perspective on memory consolidation, which is a process that transforms short-term memories into long-term ones, in AI systems.

…

A November 28 (?), 2023 IBS press release (also on EurekAlert but published December 18, 2023, which originated the news item, describes how the team went about its research,

In the race towards developing Artificial General Intelligence (AGI), with influential entities like OpenAI and Google DeepMind leading the way, understanding and replicating human-like intelligence has become an important research interest. Central to these technological advancements is the Transformer model [Figure 1], whose fundamental principles are now being explored in new depth.

The key to powerful AI systems is grasping how they learn and remember information. The team applied principles of human brain learning, specifically concentrating on memory consolidation through the NMDA receptor in the hippocampus, to AI models.

The NMDA receptor is like a smart door in your brain that facilitates learning and memory formation. When a brain chemical called glutamate is present, the nerve cell undergoes excitation. On the other hand, a magnesium ion acts as a small gatekeeper blocking the door. Only when this ionic gatekeeper steps aside, substances are allowed to flow into the cell. This is the process that allows the brain to create and keep memories, and the gatekeeper’s (the magnesium ion) role in the whole process is quite specific.

The team made a fascinating discovery: the Transformer model seems to use a gatekeeping process similar to the brain’s NMDA receptor [see Figure 1]. This revelation led the researchers to investigate if the Transformer’s memory consolidation can be controlled by a mechanism similar to the NMDA receptor’s gating process.

In the animal brain, a low magnesium level is known to weaken memory function. The researchers found that long-term memory in Transformer can be improved by mimicking the NMDA receptor. Just like in the brain, where changing magnesium levels affect memory strength, tweaking the Transformer’s parameters to reflect the gating action of the NMDA receptor led to enhanced memory in the AI model. This breakthrough finding suggests that how AI models learn can be explained with established knowledge in neuroscience.

C. Justin LEE, who is a neuroscientist director at the institute, said, “This research makes a crucial step in advancing AI and neuroscience. It allows us to delve deeper into the brain’s operating principles and develop more advanced AI systems based on these insights.”

CHA Meeyoung, who is a data scientist in the team and at KAIST [Korea Advanced Institute of Science and Technology], notes, “The human brain is remarkable in how it operates with minimal energy, unlike the large AI models that need immense resources. Our work opens up new possibilities for low-cost, high-performance AI systems that learn and remember information like humans.”

What sets this study apart is its initiative to incorporate brain-inspired nonlinearity into an AI construct, signifying a significant advancement in simulating human-like memory consolidation. The convergence of human cognitive mechanisms and AI design not only holds promise for creating low-cost, high-performance AI systems but also provides valuable insights into the workings of the brain through AI models.

Fig. 1: (a) Diagram illustrating the ion channel activity in post-synaptic neurons. AMPA receptors are involved in the activation of post-synaptic neurons, while NMDA receptors are blocked by magnesium ions (Mg²⁺) but induce synaptic plasticity through the influx of calcium ions (Ca²⁺) when the post-synaptic neuron is sufficiently activated. (b) Flow diagram representing the computational process within the Transformer AI model. Information is processed sequentially through stages such as feed-forward layers, layer normalization, and self-attention layers. The graph depicting the current-voltage relationship of the NMDA receptors is very similar to the nonlinearity of the feed-forward layer. The input-output graph, based on the concentration of magnesium (α), shows the changes in the nonlinearity of the NMDA receptors. Courtesy: IBS

This research was presented at the 37th Conference on Neural Information Processing Systems (NeurIPS 2023) before being published in the proceedings, I found a PDF of the presentation and an early online copy of the paper before locating the paper in the published proceedings.

PDF of presentation: Transformer as a hippocampal memory consolidation model based on NMDAR-inspired nonlinearity

PDF copy of paper:

Transformer as a hippocampal memory consolidation model based on NMDAR-inspired nonlinearity by Dong-Kyum Kim, Jea Kwon, Meeyoung Cha, C. Justin Lee.

This paper was made available on OpenReview.net:

OpenReview is a platform for open peer review, open publishing, open access, open discussion, open recommendations, open directory, open API and open source.

It’s not clear to me if this paper is finalized or not and I don’t know if its presence on OpenReview constitutes publication.

Finally, the paper published in the proceedings,

Transformer as a hippocampal memory consolidation model based on NMDAR-inspired nonlinearity by Dong Kyum Kim, Jea Kwon, Meeyoung Cha, C. Justin Lee. Part of Advances in Neural Information Processing Systems 36 (NeurIPS 2023) Main Conference Track

This link will take you to the abstract, access the paper by clicking on the Paper tab.

What is a multiregional brain-on-a-chip?

Leave a reply

In response to having created a multiregional brain-on-a-chip, there’s an explanation from the team at Harvard University (which answers my question) in a Jan. 13, 2017 Harvard John A. Paulson School of Engineering and Applied Sciences news release (also on EurekAlert) by Leah Burrows,

Harvard University researchers have developed a multiregional brain-on-a-chip that models the connectivity between three distinct regions of the brain. The in vitro model was used to extensively characterize the differences between neurons from different regions of the brain and to mimic the system’s connectivity.

…

“The brain is so much more than individual neurons,” said Ben Maoz, co-first author of the paper and postdoctoral fellow in the Disease Biophysics Group in the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS). “It’s about the different types of cells and the connectivity between different regions of the brain. When modeling the brain, you need to be able to recapitulate that connectivity because there are many different diseases that attack those connections.”

“Roughly twenty-six percent of the US healthcare budget is spent on neurological and psychiatric disorders,” said Kit Parker, the Tarr Family Professor of Bioengineering and Applied Physics Building at SEAS and Core Faculty Member of the Wyss Institute for Biologically Inspired Engineering at Harvard University. “Tools to support the development of therapeutics to alleviate the suffering of these patients is not only the human thing to do, it is the best means of reducing this cost.”

Researchers from the Disease Biophysics Group at SEAS and the Wyss Institute modeled three regions of the brain most affected by schizophrenia — the amygdala, hippocampus and prefrontal cortex.

They began by characterizing the cell composition, protein expression, metabolism, and electrical activity of neurons from each region in vitro.

“It’s no surprise that neurons in distinct regions of the brain are different but it is surprising just how different they are,” said Stephanie Dauth, co-first author of the paper and former postdoctoral fellow in the Disease Biophysics Group. “We found that the cell-type ratio, the metabolism, the protein expression and the electrical activity all differ between regions in vitro. This shows that it does make a difference which brain region’s neurons you’re working with.”

Next, the team looked at how these neurons change when they’re communicating with one another. To do that, they cultured cells from each region independently and then let the cells establish connections via guided pathways embedded in the chip.

The researchers then measured cell composition and electrical activity again and found that the cells dramatically changed when they were in contact with neurons from different regions.

“When the cells are communicating with other regions, the cellular composition of the culture changes, the electrophysiology changes, all these inherent properties of the neurons change,” said Maoz. “This shows how important it is to implement different brain regions into in vitro models, especially when studying how neurological diseases impact connected regions of the brain.”

To demonstrate the chip’s efficacy in modeling disease, the team doped different regions of the brain with the drug Phencyclidine hydrochloride — commonly known as PCP — which simulates schizophrenia. The brain-on-a-chip allowed the researchers for the first time to look at both the drug’s impact on the individual regions as well as its downstream effect on the interconnected regions in vitro.

The brain-on-a-chip could be useful for studying any number of neurological and psychiatric diseases, including drug addiction, post traumatic stress disorder, and traumatic brain injury.

“To date, the Connectome project has not recognized all of the networks in the brain,” said Parker. “In our studies, we are showing that the extracellular matrix network is an important part of distinguishing different brain regions and that, subsequently, physiological and pathophysiological processes in these brain regions are unique. This advance will not only enable the development of therapeutics, but fundamental insights as to how we think, feel, and survive.”

Here’s an image from the researchers,

Caption: Image of the in vitro model showing three distinct regions of the brain connected by axons. Credit: Disease Biophysics Group/Harvard University

Here’s a link to and a citation for the paper,

Neurons derived from different brain regions are inherently different in vitro: A novel multiregional brain-on-a-chip by Stephanie Dauth, Ben M Maoz, Sean P Sheehy, Matthew A Hemphill, Tara Murty, Mary Kate Macedonia, Angie M Greer, Bogdan Budnik, Kevin Kit Parker. Journal of Neurophysiology Published 28 December 2016 Vol. no. [?] , DOI: 10.1152/jn.00575.2016

This paper is behind a paywall and they haven’t included the vol. no. in the citation I’ve found.