Chronos: Learning the Language of Time Series with Abdul Fatir Ansari - #685
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
00:00
Performances of Crunos Models in Benchmark
The zero-shot crunos model, especially the best performing large model, demonstrated performance on par with task-specific models trained on the same data sets, despite never having seen these data sets during training. While crunos may not be the top performer in benchmarks like probabilistic modeling compared to models like DFT, the performance gap is fairly close. In comparisons with other time series modeling models like Moirai, Laglama, and LLM time, crunos significantly outperforms them, even in zero-shot benchmarks, showcasing its impressive performance.
Play episode from 28:06
Transcript
