【论文笔记】On the Principles of Parismony and Self-Consistency for the Emergence of Intelligence
Motivation
智能体必须在复杂环境中生存并工作,并且不得不快速且巧妙地学习反映历史经验和当前所感知的环境模型;模型被收集信息、做出决策和采取行动作为关键。
Studies within the field of neuroscience indicate that the brain's world model exhibits intricate anatomical structures, such as distinct functional modules organized hierarchically. These structures are characterized by their ability to adopt efficient neural codes for information representation, with notable examples including The Olshausen-Field model (1996) for sparse coding and representational subspaces identified through advanced computational techniques like those proposed by Chang and Tsao (2017) and Bao et al. (2020).
What to learn and how to learn.
Convolutional Sparse Coding
x=A(z)=\Sigma_{c=1}^{C}(\alpha_{1c}z_c,...,\alpha_{Mc}z_c)
核A为待优化参数。
思考
我想历史经验应该包括基因遗传。植物、动物的生化机理,和动物的先天遗传或本能,这都是生物进化、学习的结果,并被记录到DNA作遗传;另一方面,历史经验也包括幼崽从周围,尤其是向母亲模仿学习得来的模型。
自己对遗传基因如何实现为智能的认知十分有限。
我想动物的智能包含潜意识、情绪、感情、本能等,人类则在动物智能的基础上,进化出来理智、想象力、良知与道德感、信念与意志等等,我认为这些都可以称为模型。
动物与环境的互动大部分靠潜意识和本能,与自然界的周期匹配,久经考验,堪称大自然的杰作;人类由于生物体的遗传约束,尤其是神经系统的低耗能、高能效比的特点,情绪与本能往往超越理智与良知,而群体(社会)的动态演变,复杂性远超越自然,致使个体与群体在某些场景、阶段表现不尽人意,同时个体也在不断学习历史经验与教训,群体演变中有一个迟滞的反馈。
