Summary of Amend: a Mixture Of Experts Framework For Long-tailed Trajectory Prediction, by Ray Coden Mercurius et al.
AMEND: A Mixture of Experts Framework for Long-tailed Trajectory Prediction
by Ray Coden Mercurius, Ehsan Ahmadi, Soheil Mohamad Alizadeh Shabestary, Amir Rasouli
First submitted to arxiv on: 13 Feb 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG); 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 A novel modular framework for predicting pedestrians’ future motions is introduced, addressing the issue of imbalanced datasets in intelligent driving systems. The existing naturalistic trajectory prediction datasets are often skewed towards simpler samples and lack challenging scenarios, leading to underperformance on safety-critical scenarios. To tackle this problem, previous methods employed contrastive learning and class-conditioned hypernetworks, but these approaches are not modular and cannot be applied to various machine learning architectures. In this work, a specialized mixture of experts is proposed, where each expert is trained with a specific skill for a particular part of the data. A router network generates relative confidence scores to select the best expert for prediction. Experimental results on common pedestrian trajectory prediction datasets show improved performance on long-tail scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Predicting what pedestrians will do next is crucial for self-driving cars. Right now, we don’t have enough diverse and challenging training data for this task. This makes it hard for computers to make good predictions. Researchers tried to fix this problem by using special techniques like contrastive learning and class-conditioned hypernetworks. However, these methods aren’t easy to use with different types of machine learning models. In this new work, the team proposes a way to predict pedestrian movements that’s modular and can be used with many different machine learning models. They use a type of model called a mixture of experts, where each expert is trained to do well on a specific part of the data. To make predictions, they use a special network that chooses which expert to use based on how confident it is in its prediction. The results show that this new method works better than before, especially for harder-to-predict scenarios. |
Keywords
* Artificial intelligence * Machine learning * Mixture of experts
