Center for Cognitive Brain Imaging

at Carnegie Mellon University


The Center for Cognitive Brain Imaging strives for scientific innovation in understanding how the human mind works. The Center has been applying computational techniques, particularly machine learning, to fMRI brain imaging data, making it possible for the first time to relate patterns of brain activity to specific thoughts. This research attempts to characterize the nature of neural representations of a wide range of different types of concepts, and eventually, to develop a neurally-based ontology of knowledge. This general approach to understanding the brain organization for representing the meaning of concepts is called Neurosemantics.

An early step of this research approach consisted of identifying and analyzing the neural signatures of concrete concepts, like the thought of an apple or a hammer. The research progressed to identifying the experience of emotions, making it possible to tell whether someone was feeling happiness or disgust, for example. The scientific significance is that we are beginning to understand the basic neural building blocks of more and more types of concepts, from the thought of an apple to the thoughts of abstract physics concepts such as dark matter.

The approach extends to the representation of sentences and short passages, which are structured compositions of multiple concepts. The individual concepts in a sentence can be decoded from the sentence’s or passage’s fMRI signature, and the sentence structure in which they are embedded can also be decoded. The same principles apply to the representation of short passages, whose knowledge domain can be accurately decoded.

One of the most striking outcomes of this research is the very high commonality across people and across languages in how almost all types of concepts are neurally represented. A computer program (a classifier) that is trained on the fMRI data of one set of participants can then accurately decode the concepts of new participants.

Altered neural representations in suicidal ideation. Since many psychiatric conditions involve the distortion of certain types of concepts, it is possible to use neurosemantic approaches to measure the distortions and diagnose certain psychiatric conditions. For example, a person with suicidal ideation can be accurately identified by the distortions of their neural representations of concepts related to positive and negative aspects of life.

Instructional/Educational implications. Knowing the neural end-state of a domain expert in a field like physics might enable the design of an instructional program that optimally provides and assembles the building blocks that compose a targeted concept. The investigation of physics concepts representations reveals how relatively recent physics concepts (formalized only in the last few centuries) are organized in the millennia-old information system of the human brain. Furthermore, knowing the semantic content of a new science concept makes it possible to predict what its neural signature will be.

Summary.This general research approach is in its infancy, but it is advancing rapidly and could revolutionize what we can do with our brains to make them healthy and productive. Several relevant publications are available on the Publications page under Neurosemantics.