Universidad Nacional Autónoma de México, IIMAS, 04510, Mexico City, Mexico.
Universidad de Guadalajara, SUV, 44130, Guadalajara, Mexico.
Sci Rep. 2021 Mar 25;11(1):6948. doi: 10.1038/s41598-021-86270-7.
Natural memories are associative, declarative and distributed, and memory retrieval is a constructive operation. In addition, cues of objects that are not contained in the memory are rejected directly. Symbolic computing memories resemble natural memories in their declarative character, and information can be stored and recovered explicitly; however, they are reproductive rather than constructive, and lack the associative and distributed properties. Sub-symbolic memories developed within the connectionist or artificial neural networks paradigm are associative and distributed, but lack the declarative property, the capability of rejecting objects that are not included in the memory, and memory retrieval is also reproductive. In this paper we present a memory model that sustains the five properties of natural memories. We use Relational-Indeterminate Computing to model associative memory registers that hold distributed representations of individual objects. This mode of computing has an intrinsic computing entropy which measures the indeterminacy of representations. This parameter determines the operational characteristics of the memory. Associative registers are embedded in an architecture that maps concrete images expressed in modality specific buffers into abstract representations and vice versa. The framework has been used to model a visual memory holding the representations of hand-written digits. The system has been tested with a set of memory recognition and retrieval experiments with complete and severely occluded images. The results show that there is a range of entropy values, not too low and not too high, in which associative memory registers have a satisfactory performance. The experiments were implemented in a simulation using a standard computer with a GPU, but a parallel architecture may be built where the memory operations would take a very reduced number of computing steps.
自然记忆是联想的、陈述性的和分布式的，记忆检索是一种建构性的操作。此外，直接拒绝记忆中不包含的对象提示。符号计算记忆在其陈述性特征上类似于自然记忆，可以显式存储和恢复信息；然而，它们是复制性的而不是建构性的，缺乏联想性和分布式特征。在连接主义或人工神经网络范式内开发的子符号记忆是联想性的和分布式的，但缺乏陈述性属性，无法拒绝记忆中不包含的对象，并且记忆检索也是复制性的。在本文中，我们提出了一种记忆模型，该模型支持自然记忆的五个属性。我们使用关系不确定计算来模拟关联记忆寄存器，这些寄存器持有个体对象的分布式表示。这种计算模式具有内在的计算熵，用于衡量表示的不确定性。这个参数决定了记忆的操作特性。关联寄存器被嵌入到一个架构中，该架构将模态特定缓冲区中表达的具体图像映射到抽象表示中，反之亦然。该框架已被用于模拟一个手写数字的表示的视觉记忆。该系统已经使用一组带有完整和严重遮挡图像的记忆识别和检索实验进行了测试。结果表明，在关联记忆寄存器具有令人满意的性能的范围内，存在一个不是太低也不是太高的熵值范围。实验是在使用带有 GPU 的标准计算机的模拟中实现的，但是可以构建一个并行架构，其中记忆操作将只需要非常少的计算步骤。
