The thirty fifth version of NeurIPS (Neural Information Processing Systems), one of many world’s most prestigious business and educational gatherings was not too long ago concluded. NeurIPS 2021 acquired 9,122 submissions, of which 2,344 had been accepted. Twenty-six per cent of papers had been accepted (with 3 per cent designated as highlight papers), a slight improve from final yr and the very best since 2013.
One of the highlights of this yr’s convention was the introduction of a brand new award class – Dataset and Benchmark monitor. Under this class, two papers had been awarded. Machine Learning Developers Summit 2022. Last Day To Book Early Bird Passes>>
Idea behind asserting a brand new class
NeurIPS wrote in a weblog that the Datasets and Benchmarks monitor would act as a novel venue for high-quality publications and talks on pertinent subjects of precious ML datasets and benchmarks. It would additionally function a discussion board for discussions on methods to enhance dataset growth. Datasets and benchmarks are essential for the event of machine studying strategies however require their very own reviewing tips. They additionally require further particular checks like a correct description of the collected information on parameters like accessibility and bias. The submission to this monitor was reviewed based on a set of standards that had been designed particularly for datasets and benchmarks.
The following two papers had been recognised within the new class of Datasets & Benchmarks Best Paper Awards:
Reduced, Reused and Recycled: The Life of a Dataset in Machine Learning Research
This paper was revealed by a bunch of researchers from the University of California, Los Angeles, and Google Research. This paper explored the usage of datasets inside totally different machine studying subcommunities and the interplay between dataset adoption and creation. It requires researchers to pick out benchmark datasets with larger care and promote the creation of recent and extra various datasets.
This paper discovered that regardless of the foundational position of benchmarking practices in ML analysis, little consideration has been paid to benchmark dataset use and reuse dynamics. The researchers studied how the utilization patterns differ throughout ML subcommunities between 2015-2020. They discovered that the growing focus on fewer datasets inside activity communities, adoption of datasets from different duties, and focus throughout the sector on datasets which have been launched by researchers located inside a small variety of elite establishments,” the scientists famous. The results of this research can be utilized for scientific analysis, AI ethics, and fairness/entry throughout the subject.
See Also
ATOM3D Tasks on Molecules in Three Dimensions
The ATOM3D database accommodates datasets that describe the three-dimensional construction of biomolecules, together with proteins, small molecules, and nucleic acids. They characterize a wide range of essential structural, useful, and engineering challenges and function a benchmark for machine studying strategies that function on molecular construction. A Python package deal can also be supplied with all datasets, together with processing code, utilities, fashions, and information loaders for widespread machine studying frameworks akin to PyTorch. ATOM3D’s datasets are up to date as the sector progresses, and duties are added based on the mission’s wants.
At the second, Atom3D accommodates eight datasets, which may roughly be categorised into 4 sections that cowl a variety of issues starting from single molecular buildings to interactions between biomolecules and molecular useful properties and design/engineering duties.
Sohini Das
Sohini graduated from the University of Kalyani with a grasp’s diploma in nanosciences and nanotechnology. She hopes to turn out to be a tech journalist at some point. Her work focuses on digital transformation, geopolitics, and rising applied sciences.
