Summary of Bi-mamba+: Bidirectional Mamba For Time Series Forecasting, by Aobo Liang et al.
Bi-Mamba+: Bidirectional Mamba for Time Series Forecasting
by Aobo Liang, Xingguo Jiang, Yan Sun, Xiaohou Shi, Ke Li
First submitted to arxiv on: 24 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 The proposed Mamba+ block, an extension of the original Mamba state space model, aims to improve long-term time series forecasting (LTSF) by selectively combining new features with historical ones. The addition of a forget gate allows for the preservation of past information over longer periods. Furthermore, Bi-Mamba+, which applies Mamba+ in both forward and backward directions, is designed to capture interactions among time series elements. A series-relation-aware decider is also proposed to control tokenization strategies based on dataset characteristics. Experimental results on 8 real-world datasets show that the model outperforms state-of-the-art methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary LTSF helps predict future trends and patterns over a longer period. Recently, deep learning models like Transformers have shown great performance in LTSF tasks. However, they face challenges like capturing long-term dependencies and semantic characteristics. To address this, researchers proposed the Mamba model, which balances prediction performance with computational efficiency. The new Mamba+ block adds a forget gate to selectively combine historical features with new ones. This improves the model’s ability to preserve past information. Additionally, Bi-Mamba+, which applies Mamba+ in both forward and backward directions, captures interactions among time series elements. A decider is also proposed to control tokenization strategies based on dataset characteristics. |
Keywords
» Artificial intelligence » Deep learning » Time series » Tokenization
