Language Models Do Not Follow Occam's Razor: A Benchmark for Inductive and Abductive Reasoning

Reasoning is a core capability in artificial intelligence systems, for which large language models (LLMs) have recently shown remarkable progress. However, most work focuses exclusively on deductive reasoning, which is problematic since other types of reasoning are also essential in solving real-world problems, and they are less explored. This work focuses on evaluating LLMs' inductive and abductive reasoning capabilities. We introduce a programmable and synthetic dataset, InAbHyD (pronounced in-a-bid), where each reasoning example consists of an incomplete world model and a set of observations. The task for the intelligent agent is to produce hypotheses to explain observations under the incomplete world model to solve each reasoning example. We propose a new metric to evaluate the quality of hypotheses based on Occam's Razor. We evaluate and analyze some state-of-the-art LLMs. Our analysis shows that LLMs can perform inductive and abductive reasoning in simple scenarios, but struggle with complex world models and producing high-quality hypotheses, even with popular reasoning-enhancing techniques such as in-context learning and RLVR.

Semantic Softmax Loss for Zero-Shot Learning

A typical pipeline for Zero-Shot Learning (ZSL) is to integrate the visual features and the class semantic descriptors into a multimodal framework with a linear or bilinear model. However, the visual features and the class semantic descriptors locate in different structural spaces, a linear or bilinear model can not capture the semantic interactions between different modalities well. In this letter, we propose a nonlinear approach to impose ZSL as a multi-class classification problem via a Semantic Softmax Loss by embedding the class semantic descriptors into the softmax layer of multi-class classification network. To narrow the structural differences between the visual features and semantic descriptors, we further use an L2 normalization constraint to the differences between the visual features and visual prototypes reconstructed with the semantic descriptors. The results on three benchmark datasets, i.e., AwA, CUB and SUN demonstrate the proposed approach can boost the performances steadily and achieve the state-of-the-art performance for both zero-shot classification and zero-shot retrieval.