HyperFast: Instant Classification for Tabular Data

Training deep learning models and performing hyperparameter tuning can becomputationally demanding and time-consuming. Meanwhile, traditional machinelearning methods like gradient-boosting algorithms remain the preferred choicefor most tabular data applications, while neural network alternatives requireextensive hyperparameter tuning or work only in toy datasets under limitedsettings. In this paper, we introduce HyperFast, a meta-trained hypernetworkdesigned for instant classification of tabular data in a single forward pass.HyperFast generates a task-specific neural network tailored to an unseendataset that can be directly used for classification inference, removing theneed for training a model. We report extensive experiments with OpenML andgenomic data, comparing HyperFast to competing tabular data neural networks,traditional ML methods, AutoML systems, and boosting machines. HyperFast showshighly competitive results, while being significantly faster. Additionally, ourapproach demonstrates robust adaptability across a variety of classificationtasks with little to no fine-tuning, positioning HyperFast as a strong solutionfor numerous applications and rapid model deployment. HyperFast introduces apromising paradigm for fast classification, with the potential to substantiallydecrease the computational burden of deep learning. Our code, which offers ascikit-learn-like interface, along with the trained HyperFast model, can befound at https://github.com/AI-sandbox/HyperFast.

Further reading#

• Access Paper in arXiv.org