Impulse Pattern Formulation (IPF) Brain Model
원문 초록
A new brain model is introduced, based on the Impulse Pattern Formulation (IPF) already established for modeling and understanding musical instrument and rhythm perception and production. It assumes the brain works with impulses, neural bursts, ejected from an arbitrary reference point in the brain, arriving at other reflecting brain regions, and returning to the reference point delayed and damped. A plasticity model is suggested to adjust reflection strength in time. The model is systematically studied with 50 reflection points by varying the amount of excitatory vs. inhibitory neurons, the presence or absence of plasticity or external sensory input, and the strength of the input and plasticity in terms of system adaptation to an input or to the system itself. The Brain IPF shows adaptation to an external stimulus, which is stronger without plasticity, showing the active brain not being a simple passive \emph{tabula rasa}. A relation of 10-20\% of inhibitory vs. excitatory neurons, as found in the brain, shows a maximum adaptation to an external stimulus compared to all other relations, pointing to an optimum of this relation concerning adaptation. When assuming strong brain periodicities only up to about 100 Hz, the reflection strength of the model is highest for delays of around 300 ms, corresponding to Event-Related Potential (ERP) timescales of brain potentials most often found roughly between 100 - 400 ms. The mean convergence times of the model correspond to short-time memory time scales with a mean of five seconds for converging IPFs. The Brain IPF is computationally very cheap, highly flexible, and with musical instruments already found to be of high predictive precision. Therefore, in future studies, the Brain IPF might be a model able to understand very large systems composed of an ensemble of brains as well as cultural artifacts and ecological entities.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
TabPFN: One Model to Rule Them All?
미분 가능한 비선형 모델 예측 제어
Differentiable Nonlinear Model Predictive Control
매개변수적 해 민감도의 효율적인 계산은 학습 강화 방법을 비선형 모델 예측 제어와 통합하는 데 있어 핵심 과제이며, 이는 많은 학습 알고리즘에서 민감도의 가용성이 필수적이기 때문이다. 본 논문은 음함수 정리와 내점법(IPM)에서 다루는 평활화된 최적성 조건을 이용하여 일반 비선형 계획법의 해 민감도를 계산하는 방법을 논한다. 장벽 매개변수가 지정된 비음수 목표값에 접근하도록 허용하여 SQP와 IPM 이차계획법(QP) 풀이기를 일반화하고, 이 목표값이 0일 때 고전적 SQP를 복원하는 순차 장벽 이차계획법(sequential barrier quadratic programming, SBQP)을 도입한다. 기존 연구는 볼록 또는 제약 없는 정식화에 한정되거나 효율적인 민감도 평가를 위한 구현이 부족하다. 본 발표는 `acados` 프레임워크 내의 효율적인 오픈소스 구현을 동반하며, 일반 최적 제어 문제에 대해 순방향 및 수반(adjoint) 민감도를 모두 제공하고 최신 풀이기인 `mpc.pytorch` 및 `cvxpygen` 대비 3배를 넘는 속도 향상을 달성한다.
Making Language Model a Hierarchical Classifier and Generator
Implementing Adaptations for Vision AutoRegressive Model
Graph World Model
Compress Any Segment Anything Model (SAM)
토론
이 논문과 연결된 포럼 글이 아직 없습니다.