Summary of Dectrain: Deciding When to Train a Monocular Depth Dnn Online, by Zih-sing Fu et al.
DecTrain: Deciding When to Train a Monocular Depth DNN Online
by Zih-Sing Fu, Soumya Sudhakar, Sertac Karaman, Vivienne Sze
First submitted to arxiv on: 3 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Robotics (cs.RO)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes DecTrain, an algorithm that decides when to train a monocular depth neural network (DNN) online using self-supervision. The goal is to improve the DNN’s accuracy in real-world scenarios where deployment data differs from training data. DecTrain compares the cost of training with the predicted accuracy gain at each timestep and trains only 44% of the time on average, achieving similar results as online training at all timesteps while reducing computation. The algorithm is evaluated on out-of-distribution data and compared to a more generalizable high inference cost DNN and an even smaller DNN, showing that DecTrain can recover most of the accuracy gain with reduced computational costs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how neural networks (computer programs) get worse when they’re used in real-life situations. The problem is that these networks are trained on one set of data but then have to work with different data in the real world. To solve this, the researchers created a new algorithm called DecTrain. This algorithm decides when to update the network’s knowledge based on how well it thinks it will do in the future. It turns out that using this algorithm can make the network perform just as well as if it had learned from all of the data, but with much less effort. |
Keywords
» Artificial intelligence » Inference » Neural network
