MEGA AI
AI Algorithm Platform
In response to the needs of industries such as drug discovery, compound synthesis and surface defect detection of precision industrial products, MEGAROBO has built MEGA AI algorithm platform which integrates CV and NLP algorithms including advanced technologies such as small sample learning, reinforcement learning and graph calculation so as to effectively solve industry pain points with industrial-level high-precision deep learning algorithms.
TECHNICAL ARCHITECTURE
Four core modules
-
Substantial CV Algorithm Library -
Substantial NLP Algorithm Library -
Visualization of Training and Diagramming of Reasoning -
Open Super Parameter Adjustment
Substantial CV Algorithm Library
MEGA AI integrates over 10 deep learning algorithms in CV field, including improved and fine-tuned ResNet series, MobileNet series, YOLO series, YOLOX, FCOS, PFENet, UNet++, SOLO series, ABCNet, ViT series, etc.
MEGA AI integrates 30+ pre-training models for industrial scenarios, including display panel particle positioning model, display panel glass defect detection model, chip surface defect detection model, chip surface character detection and recognition model, wafer surface defect detection and segmentation model, cell segmentation model, cell tracking model, drug activity analysis, classification and clustering model.
Classification
Detection 1
Detection 2
Segmentation
Segmentation 2
Positioning
OCR
Substantial NLP Algorithm Library
MEGA AI integrates nearly 10 deep learning algorithms in the fields of NLP, reinforcement learning and graph neural networks, such as improved and fine-tuned BERT, BM25, GAT, PG, 3N-MCTS, GCNN and Retro*.
MEGA AI integrates 10+ pre-training models for industrial scenarios, including chemical keyword extraction model, chemical synthesis path prediction model, chemical reaction condition prediction model, chemical reaction efficiency prediction model, etc.
Chemical Keyword Extraction
Chemical Synthesis Path Prediction
Chemical Reaction Condition Prediction
Chemical Reaction Efficiency Prediction
Visualization of Training and Diagramming of Reasoning
In MegaTE module, the parameter changes in the current training process are dynamically displayed in the form of broken lines and curves and the parameters are also configurable.
In MegaIE module, the reasoning results are displayed intuitively in the form of charts.
Open Super Parameter Adjustment
In MegaTE module, super parameters are opened and can be flexibly configured by users to meet customization requirements such as network model selection, hardware selection, data enhancement method, learning rate, proportion of training and verification samples, iteration termination conditions etc.
Industry applications
