ITTA offers a TensorFlow catalogue for deep learning and applied AI: Deep Learning with TensorFlow 2.0 (Keras, feed-forward networks, CNN, RNN, transfer learning) and Deep Learning with TensorFlow and Neural Networks (advanced architectures, deployment, MLOps). These courses target data scientists, ML engineers, Python developers moving to AI and data engineering profiles. Delivered in Geneva, Lausanne and interactive virtual classroom.
FILTER BY
- Domains
- Editors
- Location
- Format
- Level
- Certifying
- Confirmed training
Explore TensorFlow and deep learning with ITTA’s AI training in Switzerland. From theory to production — Geneva & Lausanne, virtual or on-site.
Deep Learning TensorFlow training in Geneva and Lausanne. 1-day course on neural networks and deep learning with TensorFlow. ITTA Switzerland.
On forums and specialist press, PyTorch has taken the spotlight on the academic research and new papers side since 2023. Yet, in industrial production, TensorFlow remains extremely present, particularly in organisations that have already invested in Keras, in models trained on TensorFlow, or that leverage the Google ecosystem (TensorFlow Serving, TensorFlow Lite for mobile and edge, TensorFlow.js for the browser, TensorFlow Extended for complete MLOps pipelines). This research vs production duality explains why a data scientist in French-speaking Switzerland benefits from knowing both frameworks and choosing based on context.
On the Swiss side, TensorFlow is widely deployed in pharma, academic research (EPFL, UNIGE, ETH), international organisations doing satellite or medical image analysis, banks for scoring and fraud models, and increasingly in generative AI pipelines combining embeddings, classical models and LLMs.
Deep Learning with TensorFlow 2.0 is the recommended entry for any Python profile moving to deep learning. The programme covers TensorFlow 2 and Keras onboarding, feed-forward networks (MLP), regularisation (dropout, batch normalisation), convolutional networks (CNN) for vision, recurrent networks (RNN, LSTM, GRU) for sequences, transfer learning, and training best practices (callbacks, learning rate scheduling, monitoring). It targets data scientists and Python developers starting deep learning.
Deep Learning with TensorFlow and Neural Networks deepens architectures (Transformers, embeddings, encoder-decoder models), hyperparameter tuning, production (TensorFlow Serving, TensorFlow Lite, TensorFlow.js), model monitoring and MLOps best practices. It targets profiles moving from prototyping to industrialisation.
The deep learning landscape has changed very fast in the past two years. TensorFlow remains the de facto standard in pharma, industry, mobile and edge (TensorFlow Lite is heavily deployed in production on millions of devices). Keras 3 brings multi-backend compatibility (TensorFlow, JAX, PyTorch), opening interesting perspectives to pool code. TensorFlow Extended (TFX) remains an end-to-end MLOps reference.
On the use case side, foundation models (LLMs, vision encoders, audio) from PyTorch now dominate research, but many are available in TensorFlow versions or via Keras. The transition to LLMs and generative AI shifts the centre of gravity without making TensorFlow obsolete: production AI pipelines frequently combine classical TensorFlow models and LLM calls, in an augmentation rather than replacement logic.
TensorFlow fits the AI and data science ecosystem of our catalogue. ITTA Artificial Intelligence regroups our full AI catalogue, including LLM training (Claude, OpenAI, Gemini) now complementing classical deep learning approaches. The data science sub-domain covers the data ecosystem (pandas, scikit-learn, models), a functional prerequisite to TensorFlow. The data science and applied AI sub-domain groups business-oriented applied AI training.
On the publisher side, Python is the obvious language prerequisite, and Anthropic or OpenAI cover the complementary LLM dimension.
Our TensorFlow audience comprises junior to mid-level data scientists moving from classical machine learning to deep learning, Python developers shifting to AI, ML engineers industrialising models, academic researchers moving from Jupyter notebook to production, and data engineering profiles maintaining TensorFlow pipelines. TensorFlow skill fits a broader path (Python data, classical ML, applied LLMs) rather than as an isolated competency.
TensorFlow or PyTorch in 2026 by context?
Pragmatic question. If you start an R&D or research project from scratch, PyTorch dominates community and new models. If you work in production on mobile, edge, embedded or a Google Cloud MLOps pipeline, TensorFlow remains the solid choice. Many teams learn both. Keras 3 multi-backend now lets you write code running on both frameworks, reducing the choice cost.
What Python and ML prerequisites for the TensorFlow 2 course?
A solid Python base (variables, functions, classes, pandas, NumPy) is required. Classical machine learning exposure (scikit-learn, model notions, train/test split, metrics) greatly eases assimilation. Without these bases, we redirect to Python and data science training upstream.
How do we move from a trained TensorFlow model to a production service?
Several paths. TensorFlow Serving exposes a model via gRPC or REST API. TensorFlow Lite converts for mobile and embedded. TensorFlow.js deploys in the browser. For complete pipelines (versioning, monitoring, retraining), TensorFlow Extended (TFX) or third-party solutions (Vertex AI, SageMaker, MLflow) take over. The advanced course covers these paths.
What is the hardware cost to seriously train on deep learning?
To start, a local GPU is not required: free or paid Google Colab is enough for first projects. For production, costs relate to cloud GPU (Google Cloud, AWS, Azure) or local machines with NVIDIA GPUs. Our sessions use Colab and demonstrate training cost management best practices.
Our TensorFlow courses are delivered by data scientists and ML engineers active on client projects in French-speaking Switzerland and internationally, covering both classical deep learning and generative AI pipelines. The pedagogical commitment is concrete: code to write from the first hours, real datasets (vision, sequences, text), models to train and evaluate, and production cases discussed from field experience.
Sessions are available in Geneva, Lausanne and interactive virtual classroom. For data teams industrialising their AI stack, in-house delivery lets you calibrate training on your infrastructure (preferred cloud, GDPR/FADP constraints, existing MLOps) and priority business cases.
To make TensorFlow training concrete, our trainers rely on anonymised case studies from client projects in French-speaking Switzerland. First case: a private bank building a fraud scoring model on transactions, based on a historical dataset of several million rows. TensorFlow trains a classification model with feature engineering, category embedding and feed-forward architecture. The model is then deployed via TensorFlow Serving behind an internal API, with periodic retraining.
Second case: an international health organisation analysing medical images for diagnostic aid. TensorFlow and Keras train a CNN on transfer learning from a pre-trained EfficientNet backbone. The challenge is to obtain sufficient precision with a relatively limited dataset, properly managing cross-validation and class imbalance. The final model is packaged in TensorFlow Lite to run on edge devices without permanent connection. Third case: a Swiss scale-up predicting demand on its e-commerce platform with sequential models (LSTM then Transformer). TensorFlow compares several architectures and selects the right precision/training cost trade-off, with the full pipeline industrialised via TFX.
Monday to Friday
8:30 AM to 6:00 PM
Tel. 058 307 73 00
ITTA
Route des jeunes 35
1227 Carouge, Suisse
Monday to Friday, from 8:30 am to 06:00 pm.
