Research trends

Transcript

1. Sprint 15 Research Trends Research Trends Machine learning and deep learning with python.
2. Table Of Contents • Use of Edge Intelligence • Deep Diving Using Convolutional Neural Networks • High-Performance NLP models • Vision Transformers • Multimodal multitasking transformers
3. • Use of Edge Intelligence The ways of obtaining and analyzing data are being transformed by edge intelligence. It moves procedures from cloud-based data storage devices to the edge. By bringing decision-making closer to the data source, EI has made data storage devices somewhat independent.
4. • Deep Diving Using Convolutional Neural Networks CNN models are widely utilized in computer vision tasks such as object recognition, face recognition, and picture recognition. But, in addition to CNNs, human visual systems can distinguish them in a variety of settings, angles, and perspectives. CNNs perform 40 percent to 50 percent worse when recognizing photos in real-world object datasets. Researchers are working hard to improve this component and make it as effective in real-world applications as possible.
5. • High-Performance NLP models: ML-based NLP is still in its early stages. However, there is yet no algorithm that will allow NLP computers to recognize the meanings of various words in various situations and behave accordingly. The use of Deep Learning to improve the efficacy of current NLP systems and allow machines to interpret client questions more quickly is a primarily researched field currently.
6. • Vision Transformers The Vision Transformer, or ViT, is an image classification model that uses a Transformer-like design to classify image patches. A picture is divided into fixed-size patches, which are then linearly embedded, position embeddings added, and the resulting vector sequence given to a conventional Transformer encoder. The traditional strategy of adding an extra learnable “classification token” to the sequence is employed to perform classification. The vision transformer’s greatest significance is that it shows that we can construct a universal model architecture that can handle any sort of input data, including text, image, audio, and video.
7. • Multimodal multitasking transformers: The goal is to create a Unified Transformer model that can learn the most important tasks in a variety of domains at the same time, such as object detection, language comprehension, and multimodal reasoning. A UniT model encodes each input modality with an encoder and generates predictions on each task with a shared decoder over the encoded input representations, followed by task-specific output heads, all based on the transformer encoder-decoder architecture. With losses from each job, the full model is jointly trained from start to finish.
8. Source : https://www.marktechpost.com/2022/01/13/top-deep-learning- trends/