Currently, the Indonesian government has changed the curriculum flow based on deep learning, which includes socialisation on the application and implementation of the new curriculum to align with the independent curriculum. Therefore, I want to share information about what deep learning actually is. Deep learning is a new field in machine learning; its core idea is to simulate the hierarchical abstraction structure of the human brain, analyse large-scale data using unsupervised methods, and uncover valuable information contained within big data. Deep learning is designed to research big data and provide a deep thinking brain for big data.
“ Twenty years from now you'll be more disappointed by the things you simply didn’t do than by the ones you did do “
Charles Dickens
According to a journal from Science Direct via its website (https://www.sciencedirect.com) on the topic of learning, it states: "In June 2012, the Google Brain project attracted public attention in The New York Times". This project was led by Andrew Ng and Jeff Dean at Stanford University, using a parallel computing platform with 16,000 CPU cores to train a deep neural network with 10 billion nodes. A large amount of data was directly loaded into the algorithm, and the data was allowed to "speak". The system automatically learned from this data.
In November 2012, Microsoft publicly demonstrated a fully automatic simultaneous interpretation system in Tianjin, China. The speaker delivered a speech in English, and the computer at the back automatically performed speech recognition, translated the speech from English to Mandarin, and synthesised Mandarin speech. The key technology supporting the system behind the scenes was also deep learning (deep learning).
from the literature, it is stated that the push and pull method, that this method shows neurons in layer l actively propagating errors to neurons in layer (l-1). This is suitable for serial implementation. There is a writing conflict problem in parallel implementation. In the pull method (right figure in Figure 7.9), neurons in layer (l-1) actively obtain errors from neurons in layer l. The pull method is difficult to implement. Due to the boundary effects of the convolution operation, we need to ensure which neurons in the feature map of the previous layer are connected to the neurons in the current layer. meaning the description above of errors made at the beginning of learning or recording past information, if an error occurs, the error will continue unless there is an encouragement to find the point of error in the cognitive and application (psychomotor) domains, there is no synchronisation, so there must be an effort to encourage rethinking by analysing past learning errors or information obtained so that it can increase the stimulus to the nerves. so that the information obtained is processed into correct information based on experience.
However, this flexibility comes at a cost. In particular, deep learning architectures require a large amount of data to outperform other approaches, are computationally expensive to train, and are not easily interpretable. In the field of education, what can be taken is the method used and the scenario flow in determining improvement steps in solving a problem so that a precise and more efficient solution is obtained. Hopefully, this review can add to our insight in understanding future problems by finding the right solution.
