Neural Network Hypothesis on Interpretation and Universality

ACL ARR 2025 February Submission4806 Authors

Abstract: The network hypothesis is theoretically positioning the superiority of the neural network: (I) The neural network is not interpretable in the combination of linear models; (II) There is one general form of the common state with the accuracy in efficiency for any neural network.els. The new shaping paradigm is more promising to unlock the unknown mechanism than the conventional mathematics in the fundamental of neural network. The neuron activity are not relevant with the meaning in solving the real complex problems. The representation is imitating the human reasoning with the occurrence of the emergent ability and hallucination.

Paper Type: Long

Research Area: Interpretability and Analysis of Models for NLP

Research Area Keywords: Hallucination, Emergent Capability, Explainable AI

Contribution Types: Position papers, Theory

Languages Studied: English, Chinese

Submission Number: 4806