Infomax Neural Joint Source-Channel Coding via Adversarial Bit Flip
Authors: Yuxuan Song, Minkai Xu, Lantao Yu, Hao Zhou, Shuo Shao, Yong Yu5834-5841
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on various real-world datasets evidence that our IABF can achieve state-of-the-art performances on both compression and error correction benchmarks and outperform the baselines by a significant margin. We conduct extensive experiments on various benchmark datasets and different tasks. Empirical evidence demonstrates the effectiveness of our methods on both reducing the distortion with finite bit-length and learning useful representation for downstream tasks. |
| Researcher Affiliation | Collaboration | 1Shanghai Jiao Tong University, 2Stanford University, 3Bytedance AI lab |
| Pseudocode | Yes | Algorithm 1 Infomax Adversarial Bits Flip |
| Open Source Code | Yes | The codebase for this work can be found at https://github.com/Minkai Xu/neural-coding-IABF. |
| Open Datasets | Yes | Binary MNIST, MNIST (Le Cun 1998), Omniglot (Lake, Salakhutdinov, and Tenenbaum 2015) and CIFAR10 (Krizhevsky, Hinton, and others 2009) are provided to demonstrate the effectiveness of our methods on compression and error correction. |
| Dataset Splits | Yes | The test result is reported according to the model with the lowest distortion on the validation set. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as GPU models, CPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions software components but does not provide specific version numbers for any libraries, frameworks, or programming languages (e.g., 'PyTorch 1.9', 'Python 3.8'). |
| Experiment Setup | Yes | where λ is the only hyperparameter and selected from a small candidate set {0.1, 0.01, 0.001} during the experiments. Following (Choi et al. 2018), we also use the multi-sample objective with K = 5: LK rec(φ, θ; x, ϵ) = x D Ey1:K padv(y|x;ϵ,θ) log 1 K K i=1 qφ(x|y(i)) |