Solutional new logo (1)

Fine-tuning (deep learning)

Deel dit
" Terug naar Woordenlijst Index

Fine-tuning in deep learning is a technique used in the field of artificial intelligence[1], specifically within machine learning[2] algorithms. Its primary function is to enhance the performance of pre-existing neural network models. This is accomplished by reusing and adjusting certain parameters within these models. It’s a form of transfer learning, where knowledge gained from one task is applied to another related task. Fine-tuning can be applied to the entire network or just a subset of layers, often adding adapters for augmentation. It is particularly effective in natural language processing for language modeling. However, it’s important to note that fine-tuning can sometimes affect a model’s robustness, requiring strategies like linear interpolation to balance performance. Various methods, including the Low-rank adaptation (LoRA) technique, offer different approaches to fine-tuning.

Terms definitions
1. artificial intelligence.
1 Artificial Intelligence (AI) refers to the field of computer science that aims to create systems capable of performing tasks that would normally require human intelligence. These tasks include reasoning, learning, planning, perception, and language understanding. AI draws from different fields including psychology, linguistics, philosophy, and neuroscience. The field is prominent in developing machine learning models and natural language processing systems. It also plays a significant role in creating virtual assistants and affective computing systems. AI applications extend across various sectors including healthcare, industry, government, and education. Despite its benefits, AI also raises ethical and societal concerns, necessitating regulatory policies. AI continues to evolve with advanced techniques such as deep learning and generative AI, offering new possibilities in various industries.
2 Artificial Intelligence, commonly known as AI, is a field of computer science dedicated to creating intelligent machines that perform tasks typically requiring human intellect. These tasks include problem-solving, recognizing speech, understanding natural language, and making decisions. AI is categorised into two types: narrow AI, which is designed to perform a specific task, like voice recognition, and general AI, which can perform any intellectual tasks a human being can do. It's a continuously evolving technology that draws from various fields including computer science, mathematics, psychology, linguistics, and neuroscience. The core concepts of AI include reasoning, knowledge representation, planning, natural language processing, and perception. AI has wide-ranging applications across numerous sectors, from healthcare and gaming to military and creativity, and its ethical considerations and challenges are pivotal to its development and implementation.
2. machine learning. Machine learning, a term coined by Arthur Samuel in 1959, is a field of study that originated from the pursuit of artificial intelligence. It employs techniques that allow computers to improve their performance over time through experience. This learning process often mimics the human cognitive process. Machine learning applies to various areas such as natural language processing, computer vision, and speech recognition. It also finds use in practical sectors like agriculture, medicine, and business for predictive analytics. Theoretical frameworks such as the Probably Approximately Correct learning and concepts like data mining and mathematical optimization form the foundation of machine learning. Specialized techniques include supervised and unsupervised learning, reinforcement learning, and dimensionality reduction, among others.

In deep learning, fine-tuning is an approach to transfer learning in which the weights of a pre-trained model are trained on new data. Fine-tuning can be done on the entire neural network, or on only a subset of its layers, in which case the layers that are not being fine-tuned are "frozen" (not updated during the backpropagation step). A model may also be augmented with "adapters" that consist of far fewer parameters than the original model, and fine-tuned in a parameter–efficient way by tuning the weights of the adapters and leaving the rest of the model's weights frozen.

For some architectures, such as convolutional neural networks, it is common to keep the earlier layers (those closest to the input layer) frozen because they capture lower-level features, while later layers often discern high-level features that can be more related to the task that the model is trained on.

Models that are pre-trained on large and general corpora are usually fine-tuned by reusing the model's parameters as a starting point and adding a task-specific layer trained from scratch. Fine-tuning the full model is common as well and often yields better results, but it is more computationally expensive.

Fine-tuning is typically accomplished with supervised learning, but there are also techniques to fine-tune a model using weak supervision. Fine-tuning can be combined with a reinforcement learning from human feedback-based objective to produce language models like ChatGPT (a fine-tuned version of GPT-3) and Sparrow.

" Terug naar Woordenlijst Index
Scroll naar boven