Free Information Center
StarCoder is an open-source large language model designed to assist with programming tasks. Developed to provide code generation and understanding capabilities, it supports multiple programming languages and aims to facilitate software development processes.
Dropout as Bayesian approximation is a technique that interprets dropout in neural networks as a form of approximate Bayesian inference. This approach allows for uncertainty estimation in deep learning models by treating dropout as a variational approximation to a probabilistic model.
Task arithmetic for LLMs refers to the method of combining multiple tasks to optimize the performance of large language models. It enhances task management and efficiency.
Randomized smoothing is a technique in machine learning used to create robust classifiers by averaging predictions over noise-perturbed inputs. It provides certified robustness guarantees against adversarial attacks by transforming a base classifier into a smoothed classifier that is more resistant to input perturbations.
Physics-informed neural networks (PINNs) are a class of deep learning models that incorporate physical laws, typically expressed as differential equations, into the training process. By embedding these constraints, PINNs can solve forward and inverse problems in scientific computing with improved accuracy and generalization.
System identification for sim-to-real involves creating models that bridge the gap between simulated environments and real-world applications, enhancing the transferability of learned behaviors.
A large language model (LLM) is an advanced type of artificial intelligence designed to understand, generate, and interact using human language. These models are trained on vast amounts of text data and utilize deep learning techniques to perform a wide range of language-related tasks.
CodeGen is a family of large-scale language models designed to generate computer code from natural language prompts. Developed to assist programmers by automating coding tasks, CodeGen leverages advanced machine learning techniques and large datasets of code to improve software development workflows.
Mamba is a computational framework designed for Bayesian inference in state space models, providing flexible tools for modeling time series data with latent variables.
Knowledge distillation is a machine learning technique where a smaller model is trained to replicate the behavior of a larger, more complex model. This process helps deploy efficient models by transferring knowledge from a ‘teacher’ model to a ‘student’ model.