Point transformer (PT)
Point Transformer (PT) is a neural network architecture designed for processing 3D point cloud data. It leverages self-attention mechanisms to capture local and global geometric relationships within unordered point sets.
Technology & Innovation
Spiking neural network
A spiking neural network (SNN) is a type of artificial neural network that more closely models the behavior of biological neurons by using discrete spike events to transmit information. Unlike traditional neural networks, SNNs process data temporally and are considered a promising approach to achieving efficient and biologically plausible computation.
Learning to adapt (meta-learning for domain shift)
Learning to adapt, or meta-learning for domain shift, refers to methods in machine learning that enable models to generalize across varying domains.
PolyCoder
PolyCoder is an open-source code generation model designed to produce programming code in multiple languages. It is notable for its training on diverse datasets and its focus on programming language generation capabilities.
SPLADE (sparse lexical and dense)
SPLADE (sparse lexical and dense) is a neural information retrieval model designed to combine the benefits of sparse lexical representations and dense embeddings. It aims to improve search effectiveness by leveraging both term-level relevance and semantic matching.
CoQA (Conversational Question Answering)
CoQA is a dataset designed to facilitate the development of conversational question answering systems. It allows computers to engage in human-like dialogue by answering questions based on provided context.
Artificial intelligence alignment
Artificial intelligence alignment refers to the challenge of ensuring that AI systems act in ways that are consistent with human values, intentions, and ethical principles. It involves designing AI behaviors that are beneficial and safe, particularly as AI systems become more autonomous and capable.
Open X-Embodiment (robotics dataset)
Open X-Embodiment is a comprehensive robotics dataset designed for advancing research in embodied AI and robotics applications.
Diffsound (discrete diffusion for audio)
Diffsound (discrete diffusion for audio) is a method in generative audio modeling that applies discrete diffusion processes to create or transform audio data. It is part of a broader class of diffusion-based generative models adapted specifically for the discrete, sequential nature of audio signals.
CIFAR-100
CIFAR-100 is a dataset widely used in machine learning and computer vision, consisting of 60,000 images across 100 classes.