Junhui Hou, Hui Liu, Jiawei Tang
We have not lifted any functions out of this paper's repositories yet, so there is nothing we have run. The repositories linked to it are listed below.
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
While deep neural networks achieve remarkable performance, their tendency to produce overconfident predictions remains a critical issue. Evidential deep learning (EDL) mitigates this by modeling predictions as a Dirichlet distribution over class probabilities, thereby explicitly quantifying epistemic uncertainty. However, we identify two fundamental limitations in conventional EDL: first, its Kullback-Leibler penalty only suppresses the evidence of negative classes, which produces excessively strong evidence and consequently degrades the model's ability to quantify uncertainty; second, it lacks a theoretical guarantee for setting the Dirichlet parameter α = e + 1. In this paper, we propose a mathematically principled framework, called Variational Inference EDL (VI-EDL). By reformulating evidential learning through the lens of variational inference, we derive an Evidence Lower Bound that prevents the evidence from growing excessively. Furthermore, we rigorously establish a generalization bound, revealing how predicted uncertainty, feature representations, and network complexity affect this bound, and proving why setting α = e + 1 minimizes it. Extensive experiments on standard visual and medical datasets demonstrate that VI-EDL achieves state-of-the-art performance, exhibiting superior performance in out-of-distribution detection, noise detection, and autonomous driving scenarios. The code is available at https://github.com/seutjw/VI-EDL.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2605.26477")
get_code_for_paper("2605.26477")
have("2605.26477")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2605.26477.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2605.26477)
Connect an agent — have() is free.