ReconBoost: Boosting Can Achieve Modality Reconcilement
Authors: Cong Hua, Qianqian Xu, Shilong Bao, Zhiyong Yang, Qingming Huang
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments over six multi-modal benchmarks speak to the efficacy of the method. |
| Researcher Affiliation | Academia | 1Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China 2School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China 3Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China 4School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China 5Key Laboratory of Big Data Mining and Knowledge Management, Chinese Academy of Sciences, Beijing, China. |
| Pseudocode | Yes | Algorithm 1: Recon Boost Algorithm |
| Open Source Code | Yes | We release the code at https: //github.com/huacong/Recon Boost. |
| Open Datasets | Yes | We conduct empirical experiments on several common multi-modal benchmarks. Specifically, AVE (Tian et al., 2018) dataset [...]; CREMA-D (Cao et al., 2014) [...]; Model Net40 (Wu et al., 2015) [...]; MOSEI (Zadeh et al., 2018), MOSI (Zadeh et al., 2016), and CH-SIMS (Yu et al., 2020) are multi-modal sentiment analysis datasets [...]. |
| Dataset Splits | Yes | The training and testing split of the dataset follows the split (Cao et al., 2014). [...] The dataset split for training and testing follows the standard protocol as described in (Wu et al., 2015). [...] The training and testing split of the dataset follows the split (Zadeh et al., 2016). [...] The training and testing split of the dataset follows the split (Zadeh et al., 2018). [...] The training and testing split of the dataset follows the split (Yu et al., 2020). |
| Hardware Specification | Yes | All experiments are conducted on a Ubuntu 20.04 LTS server equipped with Intel(R) Xeon(R) Gold 5218 CPU@2.30GHz and RTX 3090 GPUs |
| Software Dependencies | No | The paper mentions that "all models are implemented with Pytorch (Paszke et al., 2017)", but it does not specify a version number for PyTorch or any other software dependencies, such as libraries or operating system versions (beyond the general Ubuntu 20.04 LTS). |
| Experiment Setup | Yes | We adopt SGD (Robbins & Monro, 1951) as the optimizer. [...] The learning rate is 0.01 initially and multiplies 0.1 every 30 stages for the CREMA-D dataset, while multiplies 0.5 after 40 stages for the AVE dataset. For MOSEI, MOSI, and Model Net40, the learning rate is 0.01 and remains constant. [...] In AVE, CREMA-D, Model Net40, MOSEI, MOSI and SIMS, T1 is 4, 4, 4, 1, 1, 1 and T2 is 4, 4, 4, 1, 1, 1 respectively. |