Training OpenNMT: Configuring a machine translation system

The training introduces participants to the world of neural machine translation through hands-on work with the OpenNMT platform. During the workshop, participants will learn the process of building, training and implementing NMT models from scratch. The program focuses on the practical aspects of configuring and optimizing translation systems, taking into account the specifics of different language pairs and application domains.The training provides a solid theoretical and practical foundation in deep learning. During an intensive workshop, participants learn the foundations of neural networks, techniques for training them and practical applications in various domains. The program combines mathematical theory with practical implementation of models, placing special emphasis on understanding the mechanisms of deep neural networks and their effective use in real projects.

• NLP specialists working on translation systems
• ML engineers interested in language processing
• Computer linguists
• Language system developers
• Researchers in the field of NLP
• Software localization specialists
• Programmers getting started with deep learning
• Data Scientists expanding their skills
• ML engineers interested in neural networks
• Data analysts entering the field of deep learning
• Scientists and researchers from various fields
• Specialists in computer vision and natural language processing

1. OpenNMT basics
• System architecture
• Preparation of the environment
• Language data management
• Text preprocessing
2. Configuration of models
• Transformer architectures
• Tokenization and dictionaries
• Training parameters
• Validation of models
3. Training and evaluation
• Training strategies
• Quality assessment metrics
• Fine-tuning of models
• Error analysis
4. Implementation of the system
• Serving models
• API integration
• Quality monitoring
• Update models
5. Foundations of deep learning
• Neural network architecture
• Activation functions and back propagation
• Optimization and regularization
• Initialization of weights and normalization
6. Neural network architectures
• Convolutional Networks (CNN)
• Recurrent networks (RNNs)
• Transformers
• Generative networks
7. Training models
• Data preparation
• Training strategies
• Validation and evaluation
• Troubleshooting
8. Implementation and optimization
• Framework’i deep learning
• Performance optimization
• Implementing models
• Best practices

The participant will gain practical knowledge in the configuration and implementation of neural machine translation systems. He will learn to prepare and process linguistic data for training NMT models. Will learn methods for optimizing translation quality for different language pairs. Will be able to implement and manage translation systems in a production environment. Will develop the ability to evaluate and improve the quality of machine translation. Will gain knowledge of NMT best practices.The participant will gain a thorough understanding of the theoretical foundations of deep learning and practical model implementation skills. He will learn to independently design and train different types of neural networks tailored to specific tasks. He will learn methods for efficient optimization and regularization of deep learning models. Will be able to diagnose and solve typical problems that occur during training of deep networks. Will gain the ability to select the appropriate network architecture for the specific task. Will develop practical skills in implementing models in popular frameworks.

• Basic knowledge of machine learning
• Knowledge of NLP issues
• Experience working with Python
• General linguistic knowledge
• Knowledge of the basics of machine learning
• A solid mathematical foundation (linear algebra, differential calculus)
• Ability to program in Python
• Basic knowledge of statistics

• OpenNMT architecture
• Preprocessing of language data
• Text Tokenization
• Transformer architectures
• NMT model training
• Translation evaluation
• Fine-tuning of models
• Serving API
• Quality monitoring
• System upgrade
• Error analysis
• Best practices NMT
• ———-|————- Subcategory | Artificial Intelligence Training code | IT-AI-160 Duration | 3 days (24 hours) Price net/person | 3750 PLN.
• Neural network architecture
• Activation functions
• Back propagation
• Gradient optimization
• Regularization and dropout
• Convolutional networks
• Recursive networks
• Transformers
• Data preparation
• Training strategies
• Validation of models
• Implementing solutions