Skip to content

14. Nové trendy

Definice

Important recent research directions - Better memory to learn complex patterns - Deep (and even deeper) learning - Better learning and optimization algorithms in general – beyond stochastic gradient descent - Joint neural embeddings - ML controllers in reinforcement learning - Metalearning and AutoML - Generative models

Otázky

Jak funguje Neural Turing Machine nebo Differentiable Neural Computer?

Neural Turing Machine

  • Je zalozen na von neumaonove architekture
  • je vsak trenovatelny pomoci gradient descent
  • Obsahuje RAM

Differentiable Neural Computer

  • The DNC is able to perform tasks that require memory and attention mechanisms by using an external memory module and a control module that can access and manipulate the data stored in the memory. This allows it to perform tasks that are difficult or impossible for traditional artificial neural networks.
  • Made up of several components, including an input module, an output module, a memory module, and a control module.
  • The output module generates output data based on the input data and the information stored in the memory module.
  • During training, the DNC is presented with input data and the corresponding output data, and it uses this data to learn to perform the desired task.
  • The DNC is able to perform tasks that require memory and attention mechanisms by using an external memory module and a control module that can access and manipulate the data stored in the memory. This allows it to perform tasks that are difficult or impossible for traditional artificial neural networks.

K čemu se používá Faster R-CNN?

  • Faster R-CNN is a two-stage object detection algorithm that is used to identify and locate objects in images and videos.
  • In the first stage, the algorithm generates a set of region proposals, which are regions in the image that are likely to contain objects. These region proposals are generated using a technique called Selective Search.
  • In the second stage, the region proposals are passed through a convolutional neural network (CNN) that classifies the objects in the regions and refines the bounding boxes.
  • The CNN is trained on a large dataset of images and their corresponding bounding boxes, which allows it to learn to recognize and classify different types of objects.
  • Faster R-CNN is a fast and accurate object detection algorithm that has been widely used in a variety of applications. It has also been the basis for many other object detection algorithms that have been developed in recent years.

Jakým způsobem funguje generování popisku k obrázku?

Pouziva se transformer s attention mechanismem

Co se optimalizuje při posilovaném učení (reinforcement learningu)?

konkurence?

Na jakých principech fungují moderní QA systémy typu ChatGPT?

transformery