RT @FrankRHutter: The data science revolution is getting closer. TabPFN v2 is published in Nature: On tabular class….

RT @_k_sridhar: Is scaling current agent architectures the most effective way to build generalist agents that can rapidly adapt? . Introduc….

RT @alfcnz: Convolutional Differentiable Logic Gate Networks @FHKPetersen

Mamba4Cast generalizes to real-life datasets from multiple domains with different patterns and granularities, given the diversity in its training setup.

Leveraging the fast linear recurrence and hidden states of Mamba, our model predicts over the entire forecast horizon in a single forward pass, scaling efficiently in comparison to its transformer-based SoTA.

With just 27M parameters, trained on a single RTX2080Ti, Mamba4Cast performs on par with SoTA zero-shot forecasting methods like Chronos-Base (200M).

The architecture is based on Mamba, and uses dilated convolutions for a large receptive field. Trained exclusively on synthetic datasets generated using 2 priors, it generalizes to out of distribution unseen time series.

We are elated to introduce our most recent work on time-series foundation models - Mamba4Cast: Efficient Zero-Shot Time Series Forecasting with State Space Models. Authors: @o_swelam, Sathya Kamesh, @julien_siems, David Salinas, @FrankRHutter. Link:

RT @Cyborgmatt: Valve has provided an additional update to the teams/players in regards to the @DOTA2 Winter Major.

RT @kunalkamra88: *😭😭😭*