MP
Jul 1, 2022
Excellent introduction to the mechanics of Neural Networks in general, and the Keras application specifically. Alec is an outstanding teacher, I always appreciate his knowledge and enthusiasm.
SS
Jun 30, 2020
Such a wonderful and high tech course in the world and it is provided by ibm and coursera.Thank you ibm and coursera for such a opportunity.I'm glad and proud to be a part of this organization.
By Xinyue Z
•Feb 7, 2022
The content is for new comer of tf.keras OK.
When you have been in this area for a long time but still want to finish the whole specilization, pls go ahead and finish the final exam directly.
The week 2 of describing the back propagation is pretty nice, not too much detail but somehow it is using an example which is even clear at some points comparing to Prof.Ng 's course, however, wont let them competet each other
By Gopal I
•Mar 11, 2022
The course materials need to be updated to clearly show how to get to the labs. The video talks of one way but the real method is different. Although one is supposed to use the IBM Labs the staff gives direction to use google colabs. I downloaded the files to local machine and discovered the dependencies by accident.
One should not expect to learn much about Keras here. It is a basic introduction to DL and NN.
By Alexey K
•Jan 8, 2023
A very light introduction into training Neural Networks with Keras. Content is of good quality, but there is not enough of it to qualify as a course. Some weeks contain ~20 min worth of videos and a quiz. I bet all the material could have been condensed into just 1 week of video lectures and a final project week.
Final project is good, but it would be way better begin auto-graded instead of peer-graded.
By Deleted A
•Aug 14, 2023
A good course overall, with reasonably good explanations and step-by-step labs. However, the course did not really cover Recurrent Neural networks and Auto Encoders properly: the theory part was barely an overview and there were no code examples or labs related to them. It is a bit disappointing to see them in the syllabus but not covered in the course.
By Kai-Thomas K
•Sep 5, 2024
The course content is really good, because it explains a lot about the real background tech. Loved that one, also the code examples. BUT: the final exam was pretty all over the place and a big leap from what was in the course examples and notebook. Probably that should be reworked.
By Vasileios D S
•Aug 22, 2021
This course is too shallow in terms of content and too light in terms of workload, it can only serve as a stepping stone towards a more advanced course. It does accomplish its stated goal of offering an introduction to many Deep Learning concepts and Keras, but nothing beyond that.
By Hamza A I
•Mar 3, 2020
The course content is very brief not much for getting a better understanding of topics. It can be a good starting course for someone but if you are interested in getting details of topics I recommend taking the courses of deeplearning.ai
By Ash P (
•May 8, 2023
+ Learnt a few things from the course.
+ The first lab about building a neural net from scratch was good.
- Passed the whole thing in just a few hours with no prior knowledge of Neural Networks or Keras.
- Short lectures.
- Very easy multiple choice questions.
- No presenter overlay on lecture videos.
- Overly prescriptive labs.
- Awkward and unnecessary dependency on Jupyter Notebook.
- Some poorly specified questions for labs.
By Xing L
•Oct 7, 2020
Too basic and simple, especially for the different popular models. Not as good as I expected.
By Francisco B F
•Apr 19, 2024
a bit outdated
By Jbene M
•Apr 1, 2020
Very Basic
By Krzysztof R
•Feb 4, 2024
Learning content was good, but making assignments was a nigthmare, I could not set up an IBM Cloud account, becasue they could not verify my payment card, when i asked them to help me, they just responded "We have reviewed your account/transaction and will not be able to offer services. No further information will be disclosed regarding this matter. Any card authorizations will reverse within 24-72 hours depending on the issuing bank.". They ghosted me completely and did not want to help. I do not recommend. It's no wonder that with this kind of customer attitude IBM is far behind the competition today.
By Chiragkumar P
•Apr 22, 2023
One of my tasks is blocked by the team and I requested them to unblock it but no one has replied yet. It's the worst management. I have one day only to finish this course and get a certificate through my institutional email id because I am leaving the current institute on Monday. So i will be not able to complete the full course after this.
By elvan e
•Nov 23, 2023
I want to continue my course and i have deadline until 25th of Nov. It is saying the page is under maintenance. Will you pay me that If I miss the deadline because of this meaintenance?
By Biswarup D
•Feb 1, 2022
PLEASE CHANGE THE INSRUCTOR,WE NEED MORE DETAILS DISCUSSION
By Mo-Javad M
•Sep 29, 2023
so bad
By Julio T
•Mar 9, 2020
j
By Amb P M S
•Apr 19, 2025
L'apprentissage profond (ou deep learning) est une sous-catégorie de l'apprentissage automatique qui repose sur des réseaux de neurones artificiels. Ces réseaux, inspirés du fonctionnement du cerveau humain, sont utilisés pour modéliser des tâches complexes telles que la reconnaissance d'images, la traduction automatique, et la classification de texte. Keras est une bibliothèque open-source en Python qui permet de construire et d’entraîner des réseaux de neurones de manière simple et rapide. Elle sert de couche d'abstraction au-dessus de bibliothèques plus bas niveau comme TensorFlow et Theano, facilitant ainsi le prototypage rapide des modèles d'apprentissage profond. 1. Qu'est-ce qu'un réseau de neurones ? Un réseau de neurones est composé de couches qui sont formées de neurones (également appelées unités ou perceptrons). Ces neurones sont organisés en trois types de couches principales : Couche d'entrée : Recevoir les données d'entrée (par exemple, les pixels d'une image, un vecteur de texte, etc.). Couches cachées : Effectuer des transformations successives des données d'entrée via des poids et des fonctions d'activation. Couche de sortie : Produire une prédiction ou une classification à partir des données traitées. 2. Principe du fonctionnement d'un réseau de neurones : Chaque neurone reçoit une somme pondérée de ses entrées, puis applique une fonction d'activation pour produire une sortie. Par exemple, pour une couche cachée, la sortie 𝑦 y d'un neurone peut être calculée par : 𝑦 = 𝑓 ( 𝑊 ⋅ 𝑥 + 𝑏 ) y=f(W⋅x+b) Où : 𝑊 W est le vecteur des poids, 𝑥 x est le vecteur des entrées, 𝑏 b est le biais, 𝑓 f est la fonction d'activation (comme ReLU, sigmoïde, tanh, etc.). La fonction d'activation joue un rôle crucial en apportant de la non-linéarité au modèle, permettant ainsi au réseau de neurones de capturer des relations complexes entre les données. 3. Keras : Pourquoi et comment l'utiliser ? Keras est une bibliothèque qui simplifie la création, l'entraînement et l'évaluation de modèles de réseaux de neurones. Son design est très intuitif, ce qui en fait une excellente option pour les débutants. a. Installation de Keras Keras est intégré dans TensorFlow, donc si vous installez TensorFlow, vous obtenez Keras automatiquement. bash Copier Modifier pip install tensorflow b. Créer un modèle avec Keras Voici un exemple simple de création d'un réseau de neurones en utilisant Keras pour résoudre un problème de classification avec un jeu de données comme MNIST (images de chiffres manuscrits). Étapes : Importer les bibliothèques nécessaires python Copier Modifier import tensorflow as tf from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Flatten from tensorflow.keras.datasets import mnist from tensorflow.keras.utils import to_categorical Charger et préparer les données python Copier Modifier # Charger le jeu de données MNIST (x_train, y_train), (x_test, y_test) = mnist.load_data() # Normaliser les images pour qu'elles soient entre 0 et 1 x_train, x_test = x_train / 255.0, x_test / 255.0 # Convertir les labels en format one-hot (pour la classification) y_train = to_categorical(y_train, 10) y_test = to_categorical(y_test, 10) Construire le modèle python Copier Modifier # Créer un modèle séquentiel model = Sequential() # Ajouter une couche de transformation des images en vecteurs model.add(Flatten(input_shape=(28, 28))) # Ajouter une couche dense (fully connected) avec 128 neurones model.add(Dense(128, activation='relu')) # Ajouter une couche de sortie avec 10 neurones (1 par chiffre de 0 à 9) model.add(Dense(10, activation='softmax')) Compiler le modèle python Copier Modifier model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy']) Entraîner le modèle python Copier Modifier model.fit(x_train, y_train, epochs=5) Évaluer le modèle sur les données de test python Copier Modifier test_loss, test_acc = model.evaluate(x_test, y_test) print(f"Test accuracy: {test_acc}") 4. Explication des étapes clés : Chargement des données : Ici, on charge le jeu de données MNIST, qui contient des images de chiffres écrits à la main (28x28 pixels). Les labels sont les chiffres correspondants. Prétraitement des données : Les images sont normalisées pour que les valeurs des pixels soient entre 0 et 1 (cela aide le modèle à converger plus rapidement). Création du modèle : On construit un réseau de neurones avec une couche d'entrée qui a la même forme que les images (28x28), suivie d'une couche dense avec 128 neurones, et une couche de sortie avec 10 neurones pour les 10 classes possibles. Compilation du modèle : On définit l'optimiseur, la fonction de perte et les métriques. L'optimiseur Adam est souvent un bon choix pour les réseaux de neurones. Entraînement : Le modèle est entraîné sur les données d'entraînement pendant 5 époques. Évaluation : Après l'entraînement, on évalue la performance du modèle sur le jeu de test. 5. Concepts clés à comprendre : Fonction de coût (Loss function) : Elle mesure l'écart entre la sortie prédite et la vérité terrain. Par exemple, pour un problème de classification, la fonction de coût pourrait être l'entropie croisée. Optimisation (Gradient Descent) : L'optimiseur ajuste les poids du modèle pour minimiser la fonction de coût. Keras permet de choisir entre plusieurs optimisateurs comme SGD, Adam, RMSProp, etc. Époques et lots (Epochs & Batches) : Une époque correspond à un passage complet sur l'ensemble des données d'entraînement. Les données sont généralement divisées en lots (mini-batch) pour accélérer l'entraînement. 6. Améliorations possibles du modèle : Réseaux convolutifs (CNNs) : Pour des tâches comme la reconnaissance d'images, vous pouvez utiliser des réseaux de neurones convolutifs, qui sont plus efficaces pour capturer les caractéristiques spatiales des images. Réseaux récurrents (RNNs) : Si vous travaillez avec des séquences de données (comme du texte ou de la musique), vous pouvez utiliser des réseaux de neurones récurrents (LSTM, GRU). Régularisation : L'ajout de techniques comme la régularisation L2 ou Dropout peut aider à éviter le surapprentissage (overfitting). 7. Conclusion : Keras est un excellent outil pour débuter avec l'apprentissage profond grâce à sa simplicité et sa flexibilité. En quelques lignes de code, vous pouvez créer des modèles de réseaux de neurones efficaces pour des tâches variées. L'un des avantages majeurs de Keras est son intégration avec TensorFlow, ce qui vous permet de passer à des tâches plus complexes tout en profitant de la puissance de TensorFlow pour la production et l'optimisation des modèles.
By saurabh
•Feb 21, 2025
A Truly Transformative Learning Experience The "Introduction to Deep Learning & Neural Networks with Keras" course on Coursera is an absolute game-changer for anyone serious about mastering AI. This course delivers a perfect balance of theory and hands-on practice, making complex deep learning concepts incredibly intuitive. The instructor’s clarity and expertise elevate the learning experience, breaking down neural networks, activation functions, and optimization techniques with unparalleled precision. The seamless integration of Keras and TensorFlow allows for immediate practical application, reinforcing concepts through real-world projects. What sets this course apart is its structured approach—from foundational principles to implementing deep learning models with ease. Every lecture is engaging, insightful, and meticulously designed to ensure maximum retention and skill development. As someone committed to AI excellence, this course has sharpened my understanding and accelerated my journey toward becoming a top-tier AI engineer. I highly recommend it to anyone looking to build a strong foundation in deep learning.
By Marc A T
•Oct 23, 2022
I really like this course as it has only provided me with the tools that I need to understand neural networks without overhelming myself with abstract mathematics or theories. It is indeed good to be equipped with foundational theories and mathematics, but they do not build much intuition without direct practice. I believe that theories can be learned along the process rather than being stagnant because you cannot grasp them initially. With this course, I have become proactive because this is more concept-based and the mathematics has a good visualization to it. I like the way it is designed and how the final requirement aids you while allowing flexibility.
This course is highly recommended to those who have a background in Python programming, and interested in the basics of neural networks without the exhaustive mathematics.
By Frank S
•Nov 22, 2021
This was a good course and well paced. But I have a comment about the final project. It was not clear to me if the same model was being re-trained and improved upon in 50 trials - or if the idea was to do 50 trials to capture just how good a model it is, using different splits of the data. Since we were asked to find the mean and sigma of 50 trials - I assumed those trials would be independent and each one performed with a fresh model - otherwise it is odd to find a mean and sigma of a system that is converging to a final result. In that case it's only the final mean and sigma of the final trial that matters. So I think some clarity would help there - without giving it all away with regard to how to do the project.
By Ilayaperumal K
•Jan 30, 2023
Everything depends on AI & ML now. Before enrolling in Coursera’s IBM AI Engineering Professional Certificate Course, I had zero knowledge about Neural Networks & Deep learning. After completing the course, I am able to use these applications properly. It proved to be a total career game-changer.
Coursera has proven to be a wonderful experience. I have taken classes from other organizations, but there’s no comparison to Coursera.
Highly relevant examples were given which made me get a clear idea of the concepts explained. Deeply appreciate Coursera & IBM for designing this course very well.
Kudos to Instructor Alex Aklson, Ph.D., Data Scientist for providing such a wonderful learning Journey.
By shaima
•Jul 21, 2024
This course offers a thorough introduction to deep learning and neural networks with a focus on practical application using Keras. It is well-suited for anyone looking to get started with deep learning or deepen their understanding of neural networks. The combination of clear instruction, hands-on practice, and real-world examples makes it an invaluable learning resource. I highly recommend this course to anyone interested in deep learning and neural networks. Whether you are a beginner or have some prior experience, this course will provide you with the knowledge and skills necessary to excel in the field of deep learning.
By Gurpal S G
•Feb 22, 2024
IBM's "Introduction to Deep Learning & Neural Networks with Keras" course is a game-changer for anyone stepping into the realm of deep learning. From start to finish, it's a journey packed with clarity, hands-on practice, and real-world relevance. The course breaks down complex concepts into easier and digestible parts, making it perfect for beginners while still engaging for those with some background. What sets it apart is the practical approach using Keras, a user-friendly tool that simplifies deep learning implementation. I learnt so much from the final project, which was fascinating implementing machine learning!
By Kelvin M
•Jun 12, 2023
I loved the Introduction to Deep Learning & Neural Networks with Keras course! It was a great way to learn the basics of deep learning and how to use Keras, a popular and user-friendly framework for building neural networks. The course was well-structured, with clear and engaging lectures, quizzes, assignments and peer reviews. The instructor was very knowledgeable and helpful, and the practical elements of the course were very helpful. I learned a lot from this course and I highly recommend it to anyone who wants to start their journey in deep learning!