Benchmarking and Confidence Evaluation of LALMs For Temporal Reasoning

TL;DR

This paper introduces TREA, a new dataset for evaluating large audio language models' temporal reasoning abilities, and proposes an uncertainty metric to assess model confidence and robustness.

Contribution

It presents a novel dataset for temporal reasoning in audio LALMs and introduces an uncertainty metric for better evaluation of model reliability.

Findings

LALMs lag behind human performance on TREA tasks.

Uncertainty metrics do not always correlate with accuracy.

Evaluation methods need to be comprehensive for high-stakes use cases.

Abstract

The popular success of text-based large language models (LLM) has streamlined the attention of the multimodal community to combine other modalities like vision and audio along with text to achieve similar multimodal capabilities. In this quest, large audio language models (LALMs) have to be evaluated on reasoning related tasks which are different from traditional classification or generation tasks. Towards this goal, we propose a novel dataset called temporal reasoning evaluation of audio (TREA). We benchmark open-source LALMs and observe that they are consistently behind human capabilities on the tasks in the TREA dataset. While evaluating LALMs, we also propose an uncertainty metric, which computes the invariance of the model to semantically identical perturbations of the input. Our analysis shows that the accuracy and uncertainty metrics are not necessarily correlated and thus, points to a need for wholesome evaluation of LALMs for high-stakes applications.