Scholarlink

This article introduces a novel mathematical and computational framework for epistemic value calculation within deep active inference models. We focus on a visual foraging problem in a static environment, using toy and real-world MNIST datasets to explore the drawbacks and advantages of the standard method compared to our proposed approach. Testing with relevant metrics, our approach demonstrates improved results in the considered scenarios.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Epistemic Value Anticipation into the Deep Active Inference Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Nikita Fedosov,
  • Alexey Voskoboynikov,
  • Alexey Ossadtchi

摘要

This article introduces a novel mathematical and computational framework for epistemic value calculation within deep active inference models. We focus on a visual foraging problem in a static environment, using toy and real-world MNIST datasets to explore the drawbacks and advantages of the standard method compared to our proposed approach. Testing with relevant metrics, our approach demonstrates improved results in the considered scenarios.