Summary of Differentiable Interacting Multiple Model Particle Filtering, by John-joseph Brady et al.
Differentiable Interacting Multiple Model Particle Filtering
by John-Joseph Brady, Yuhui Luo, Wenwu Wang, Víctor Elvira, Yunpeng Li
First submitted to arxiv on: 1 Oct 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: Machine Learning (cs.LG); Signal Processing (eess.SP)
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 sequential Monte Carlo algorithm is proposed to learn model parameters when the underlying model exhibits random discontinuous jumps in behavior. The algorithm, based on differentiable particle filtering, trains high-dimensional parameter sets associated with neural networks using gradient descent. A new differentiable interacting multiple model particle filter is designed to learn individual behavioral regimes and the controlling model simultaneously, allowing control over computational effort per regime and guiding sampling using probability of being in a given regime. Additionally, a new gradient estimator with lower variance than established approaches is developed, proven to be consistent, and demonstrated to outperform previous state-of-the-art algorithms numerically. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates an algorithm that can learn about models that have random changes in behavior. The model uses something called differentiable particle filtering, which helps train high-dimensional parameters like those found in neural networks. The new algorithm is designed to work with multiple behavioral regimes and the controlling model, giving it more control over how much computation it does per regime. It also develops a new way to estimate gradients that has lower variance and is fast to compute. |
Keywords
* Artificial intelligence * Gradient descent * Probability
