nextup neuro
Our vision is to offer a coherent path to substrate independent human minds. By developing whole brain emulation technology through a science and engineering-oriented plan, we ensure systematic, step-wise, and accurate progress.
Our aims
Establish metrics and standardized tests useful regardless of methodology
Develop proof-of-concept tools for emulating partial and full brains
Demonstrate application of constraints and scale separation
At present
There is no proof-of-concept brain emulation.
There is no similarity metric based on success criteria for emulation.
There is no demonstrated application of constraints and scale separation.
Our first product is a challenge data set that will be made freely available to anyone, allowing anyone with a proposed method of system identification and emulation to subject that method to a standardized test with clear feedback about the degree of emulation that was achieved, a score, and identified points for improvement.
Whole brain connectome data is becoming available. There are two principal challenges to carry out system identification and to translate that data to model architecture and tuned parameters in a whole brain emulation:
At present, it is impossible to verify if an attempt to reconstruct brain tissue has achieved the intended goal. If it fails to operate as expected it can be nearly impossible to determine why or to fix it, because the system being reconstructed is not fully known. Without that knowledge, you cannot apply success criteria, you cannot measure the level of success, you cannot systematically determine how to improve your method. Nextup Neuro is working to produce fully known ground-truth systems, both in-silico and in controlled biological experimental settings, as well as to devise a set of success criteria and corresponding metrics.Â
Translating recorded brain data to a model architecture with tuned parameters by itself is an extremely difficult computational task. The problem rapidly leads to a combinatorial explosion of possibilities with a search space too large to explore, or, for which there is insufficient data to converge on a verifiably acceptable solution. Nextup Neuro is working to develop a systematic process for the application of constraints at multiple levels, commencing by learning methods with fully known ground-truth systems that we extend to brain systems.