AI-FGCT Artificial Intelligence Precision-Guided Cell Technology Recurrent Neural Network RNN Architecture
About the Recurrent Neural Network (RNN) architecture
Complete the marshalling of cell classification data in process control.
Liu Zhe
About author:Prof. Zhe Liu, academician of the European Academy of Natural Sciences, took the lead in proposing the concept of AI-FGCT. （ artificial intelligence-fine guidance stem cell technology”）。
That is, "artificial intelligence-guided stem cell technology".
There are many data that need to be considered in cytology research and industrial cell production, such as the concentration of culture fluid, the consumption of various nutrients, the maintenance and change of the physical and chemical environment, microbial contamination, iterative passage time, cell adherence and suspension, and the concentration monitoring of various nutrients and inducible substances. Although a large amount of data is ordered and set, many associations are not interactively locked and processed in an orderly manner.
Based on the neural network algorithm, the Recurrent Neural Network (RNN) architecture is generated to complete the marshalling of cell classification data in process control. The industrial requirements of cell research and cell production, although complex, are well organized with large amounts of data and processes. Process control requires processing a large amount of data grouping, and one of the major challenges in the control process is that all processes are ongoing, dynamic. And one ring after another, there is a strong correlation. When there is a technical problem, you can't shut down the entire production process, or what kind of impact will it have on the whole process after the problem, and you need to react immediately and actually, and the conventional means require a huge amount of computation, here we need to use the RNN architecture for artificial intelligence.
RNN solves the problem in two main ways: 1. Sequence data processing: RNN can handle the situation that multiple input conditions correspond to multiple data outputs in the process of cell growth, especially for sequence data, such as quantity sequences, temperature or culture liquid concentration, where each output is related to the current and previous inputs. For example, if cells grow too fast, the faster the concentration of the culture fluid is consumed. 2. Cyclic connection: Cyclic connection in RNN enables the network to capture the correlation between inputs, so as to use the previous input information to influence subsequent outputs. Once cell growth has begun, it cannot be aborted in process control.
The components of an RNN include:
Input Layer: Receives input data and passes it to the hidden layer. The input is not just static, but also contains historical information in the sequence.
Hidden layer: The core part that captures timing dependencies. The output of a hidden layer depends not only on the current input, but also on the hidden state at the previous moment.
Output Layer: Generate the final result based on the output of the hidden layer.
LSTM is a time-loop neural network that has its own advantages and disadvantages compared to RNN. Variants of LSTM networks: bidirectional recurrent neural networks and deep recurrent neural networks. The main structure of a bidirectional recurrent neural network is composed of two unidirectional recurrent neural networks. At each point t, the input can be supplied to two recurrent neural networks in opposite directions at the same time, and the output is determined by the two unidirectional recurrent neural networks. The LSTM sets six weights, corresponding to the input to the forward and backward hidden layers (W1, W3), the hidden layer to the hidden layer itself (W2, W5), and the forward and backward hidden layers to the output layer (W4, W6). In order to ensure that the unfolded graph is acyclic and there is no information flow between the forward and backward hidden layers, the deep recurrent neural network: in order to enhance the data expression feature ability of the studied model, the network replicates the cyclic structure many times at each time point, but the information is optimized. The parameters are consistent in the loop body for each layer, and the data and the number of cycles can be recorded. The parameters in different layers can have different settings.
RNN vs. LSTM.
LSTM solves this problem by introducing the concept of "cell state", which allows you to always remember previously entered information, which is a kind of selective memory, retaining important information and ignoring unimportant information.
From LSTM to Gated Recurrent Unit (GRU): Structure and Parameter Simplification: Compared with LSTM, the structure and parameters of GRU are simplified. This means that the GRU requires less storage space and is computationally faster. Improved computing efficiency: GRU is more efficient than LSTM. This is useful in scenarios where a quick response is required or where computing resources are limited.
LSTM vs. GRU.
In the process of cell production, the main function of RNN is to complete the marshalling of cell classification data in process control, and the advantage of RNN is to process the temporal relationship of instruction values of words or characters in the text, and to classify or translate the text. Process the temporal information in the voice command value signal and convert it into the corresponding editable text. and other data with time series characteristics. At the same time, it is good at processing data, processing frame sequences, and extracting key features in the setting. Break down a research or production process into a series of keyframes and generate editable text for each frame that matches the content. Presets allow you to generate a summary of typical scenarios.
Without the algorithm of artificial intelligence, the amount of data and storage is unimaginable, and the storage space and computing power consumed will not be able to ensure normal research and production analysis. In the process control of cell production, the main function of RNN is to complete the grouping of data, and carry out artificial intelligence data grouping operation according to the target requirements given by cell researchers to perform cyclic fit analysis of duplicate data.