This article presents MAMI, a method to improve long sequence neuron captioning. Neuron labeling is a way to visualize how neurons in a deep neural network respond to certain patterns. Previous work, MILAN, used an encoder-decoder image captioning approach for this, but was not effective for long sequences. MAMI improves upon this by using different attention mechanisms and combining their results. The proposed model achieved higher BLEU and F1-Score, indicating improved performance.
Publication date: 8 Jan 2024
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2401.02744
