Scientific Reports                          volume  14, Article number: 27249  (2024 )             Cite this article

Untrained neural networks can demonstrate memorization-independent abstract reasoning

submited by
Style Pass
2024-12-11 23:30:03

Scientific Reports volume  14, Article number: 27249 (2024 ) Cite this article

The nature of abstract reasoning is a matter of debate. Modern artificial neural network (ANN) models, like large language models, demonstrate impressive success when tested on abstract reasoning problems. However, it has been argued that their success reflects some form of memorization of similar problems (data contamination) rather than a general-purpose abstract reasoning capability. This concern is supported by evidence of brittleness, and the requirement of extensive training. In our study, we explored whether abstract reasoning can be achieved using the toolbox of ANNs, without prior training. Specifically, we studied an ANN model in which the weights of a naive network are optimized during the solution of the problem, using the problem data itself, rather than any prior knowledge. We tested this modeling approach on visual reasoning problems and found that it performs relatively well. Crucially, this success does not rely on memorization of similar problems. We further suggest an explanation of how it works. Finally, as problem solving is performed by changing the ANN weights, we explored the connection between problem solving and the accumulation of knowledge in the ANNs.

The topic of this paper is abstract reasoning, sometimes referred to as “fluid intelligence”1. Abstract reasoning is, broadly speaking, the ability to solve complex problems by identifying regularities and relations in the problem being solved and utilizing them for deducing the solution2,3. It is often studied using intelligence tests that comprise word analogy tests (e.g., infer that the relationship between “cow” and “milk” is the same as between “chicken” and “egg”) and visual reasoning tests (e.g., Raven Progression Matrices)4,5. As artificial intelligence continues to advance, understanding the nature of abstract reasoning in both humans and machines is becoming a central question in cognitive science and AI research6.

Leave a Comment