This project developed a new GPU-based implementation of the Potts latching model. The code was written in MATLAB and can be run on any personal computer with an NVIDIA gpu, without needing access to a large computing cluster.
Det ble utviklet en GPU-optimisert versjon av Potts-modellen, utviklet i MATLAB, som gjør det mulig å kjøre simuleringer uten tilgang til en cluster. Dette vil gjøre det enklere og billigere å drive forskningen med disse modellene i framtiden.
Prosjektlederen har utvidet kompetansen sin innenfor programmering, datamodellering, og ulike mattematiske metoder forbindet med statistisk fysikk.
Det internasjonalet samarbeidet mellom NTNU og LIMBO-gruppen ved SISSA i Trieste opplevdes som verdifullt av alle som var vitenskapelig involvert med prosjektet, og har dermed skapt presedens for lignende samarbeid i framtiden.
Konferansen som ble organisert i samspill med prosjektets slutt, førte til en viktig utbytte av kunnskap mellom forskere fra ulike land og fagfelt, og vil mest sannsynlig føre til en publisering av en proceedings som vil hjelpe til å videreføre forskningen og kunnskapen.
The goal of this project is to develop a theoretical model of one of the most robust empirical findings of cognitive neuroscience, the so-called 'N400' - an electophysiological response to semantically anomalous words and other stimuli, which can be measured at the scalp using realtively cheap and non-invasive methods. Over several decades, neuroscientists have amassed an impressive corpus of data regarding what types of stimuli affect the amplitude and other properties of our N400 measurements. Despite this, we currently have little meaningful insight into what actually causes the N400 in neural terms, or indeed why it should exist in the first place.
Our goal is to build on existing empirical and theoretical work, and construct a neural model of the N400. This way we can not only propose a theory of why the N400 exists, but also gain new insight into what our knowledge of the N400 actually tell us about the way brain categorises and processes semantic information.
Our focus is on the use of latching Potts models. These belong to a broader class of models known as "attractor networks", which represent some of the best studied models of neural function. The latching Potts model is exceptional in its ability to spontaneously and successively recall memories, dependent on the relationship between those memories. In the case that the memories represent linguistics elements (words, etc.) it can compose structures (e.g. phrases, sentences) and thereby function as a dynamical model of language. This allows us to study the N400 as a function of complex linguistic contexts.
