Rajat Sen

research scientist @ google ai, time-series (see TimesFM - llm efficiency, bandits

[Rajat Sen @rsen91]

@krasul Link: https://t.co/rhwLqOf7m8

[Rajat Sen @rsen91]

Thanks to @krasul TimesFM-2.5 is now available as a native huggingface transformers model as well. Links to follow.

[Kashif Rasul @krasul]

@JeffDean @GoogleResearch Also being added to transformers https://t.co/kWbXEunzfx

[Jeff Dean @JeffDean]

Learn more about a very high quality time series model released by @GoogleResearch a while back at https://t.co/GClBKeDPhf

[Rajat Sen @rsen91]

@0xkatienarain @ElaPassarelli +1 for brigadiers

[Rajat Sen @rsen91]

@iclr_conf we put in tons of effort for the rebuttal including many new experiments and got no replies from the reviewers. Now we got an email that the original reviews will be frozen with no chance for the reviewers to respond or change the scores. This does not seem fair!

[Ivan Kuznetsov @ivanku]

Proud to see @GoogleResearch partnership with @googlecloud BigQuery team deliver more value today! TimesFM 2.5 is now available in BQML, unlocking AI.DETECT_ANOMALIES.

Kudos to my research colleagues Abhimanyu Das, @rsen91, Yichen Zhou and our amazing Cloud partners Haoming Chen, Vaibhav Sethi.

Why it matters: It catches "silent" failures that static alerts miss. If traffic drops on a Monday (when it should be high), TimesFM spots it immediately based on historical context, even if it doesn't hit a "zero" threshold.

https://t.co/2aWnB10KCq

[Google Research @GoogleResearch]

We present a new approach to time-series forecasting that uses continued pre-training to teach a model to adapt to in-context examples at inference time, matching the performance of supervised fine-tuning without additional complex training. Learn more at https://t.co/1nVrRBaETx

[Rajat Sen @rsen91]

We just released the weights of TimesFM-2.5 which is now the best zero-shot forecasting model on Gift-Eval on all metrics.

Compared to previous versions:
-- it can be more accurate by up to 25% while being half the number of parameters
-- should have improved probabilistic forecasts
-- 4x longer context

We should have these models available on GCP via BigQuery and Model Garden soon. I am excited about the future of forecasting at Google Research -- especially about unlocking new zero-shot capabilities.

[Google Research @GoogleResearch]

Excited to release TimesFM 2.5 on Hugging Face (soon available in BigQuery and Model Garden), significantly improving over TimesFM 2.0 in accuracy and maximum context length. TimesFM 2.5 tops the GiFT-Eval leaderboard on all accuracy metrics among zero-shot foundation models.

[Rajat Sen @rsen91]

@MountainOfMoon Congratulations! Well deserved

[Rajat Sen @rsen91]

@yuqirose Congrats Rose!

[Google AI @GoogleAI]

We just released TimesFM-2.0 (jax & pytorch) on Hugging Face (https://t.co/JfcV1yOyx0), with a significant boost in accuracy and maximum context length.

TimesFM-2.0 tops the GIFT-Eval (https://t.co/C5cXGipNxQ) leaderboard on point and probabilistic forecasting accuracy metrics.

[Rajat Sen @rsen91]

@Michael_Druggan I guess one complication is that in tech (as you know already) most of the higher compensation is paid in stock and is not reflected in the base salary. The current uscis records only include the base.

[Rajat Sen @rsen91]

@AlexGDimakis Congrats Alex 🎉

[Jeff Dean @JeffDean]

What a way to celebrate one year of incredible Gemini progress -- #1🥇across the board on overall ranking, as well as on hard prompts, coding, math, instruction following, and more, including with style control on.

Thanks to the hard work of everyone in the Gemini team and elsewhere at Google! 🎊

[Peter Moskos @PeterMoskos]

Not quite. Poor people and all transit commuters yearn for public transportation without criminals and mentally ill people enabled by upper middle class apologists who think such things are synonymous with being poor.

[Rajat Sen @rsen91]

We just released a PyTorch version of the original timesfm-1.0 model (https://t.co/nIJALnGiZG), with more details at https://t.co/shcfW2K4kz

[Google AI @GoogleAI]

TimesFM is a forecasting model, pre-trained on a large time-series corpus of 100 billion real world time-points, that displays impressive zero-shot performance on a variety of public benchmarks from different domains and granularities. Learn more → https://t.co/U1OctNPZET

[Rajat Sen @rsen91]

@MountainOfMoon Many congratulations!

[Rajat Sen @rsen91]

@nikos_kafritsas Thanks, we will be releasing something to that effect soon!

[Rajat Sen @rsen91]

We released an example notebook (https://t.co/QHcUv31GfF) for fine-tuning the pre-trained TimesFM model on your own dataset. Some more exciting updates to follow!

[Google AI @GoogleAI]

TimesFM is a forecasting model, pre-trained on a large time-series corpus of 100 billion real world time-points, that displays impressive zero-shot performance on a variety of public benchmarks from different domains and granularities. Learn more → https://t.co/U1OctNPZET

[Jeff Dean @JeffDean]

Very nice results from a @GoogleAI research effort on a general purpose time-series prediction model that gives good zero-shot performance to new forecasting tasks.