diff --git a/.readthedocs.yml b/.readthedocs.yml
index 44d7f569e..7a479688a 100644
--- a/.readthedocs.yml
+++ b/.readthedocs.yml
@@ -9,7 +9,7 @@ version: 2
 build:
   os: ubuntu-22.04
   tools:
-    python: "3.8"
+    python: "3.10"
 
 # Build documentation in the docs/ directory with Sphinx
 sphinx:
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 542df512c..b5d2d37a8 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -12,6 +12,10 @@ Changelogs for this project are recorded in this file since v0.2.0.
 
 ## [Towards v0.7]
 
+### Changed
+
+* The `shapelets` module now depends on Keras3+ (should be keras-backend-blind) and not anymore on TF
+
 ## [v0.6.3]
 
 ### Changed
diff --git a/azure-pipelines.yml b/azure-pipelines.yml
index cc04f70c0..347525f62 100644
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@@ -10,8 +10,8 @@ jobs:
     vmImage: 'ubuntu-latest'
   strategy:
     matrix:
-      Python38:
-        python.version: '3.8'
+      # Python38:
+      #   python.version: '3.8'
       Python39:
         python.version: '3.9'
   variables:
@@ -42,8 +42,7 @@ jobs:
       set -xe
       python -m pip install pytest pytest-azurepipelines
       python -m pip install scikit-learn==1.2
-      python -m pip install tensorflow==2.9.0
-      python -m pytest -v tslearn/ --doctest-modules
+      export KERAS_BACKEND="torch" && python -m pytest -v tslearn/ --doctest-modules
     displayName: 'Test'
 
 - job: 'linux_without_torch'
@@ -80,9 +79,7 @@ jobs:
   - script: |
       set -xe
       python -m pip install pytest pytest-azurepipelines
-      python -m pip install scikit-learn==1.2
-      python -m pip install tensorflow==2.9.0
-      python -m pytest -v tslearn/ --doctest-modules -k 'not tslearn.metrics.softdtw_variants.soft_dtw and not tslearn.metrics.softdtw_variants.cdist_soft_dtw and not tslearn.metrics.dtw_variants.dtw or tslearn.metrics.dtw_variants.dtw_'
+      python -m pytest -v tslearn/ --doctest-modules --ignore tslearn/shapelets/ -k 'not tslearn.metrics.softdtw_variants.soft_dtw and not tslearn.metrics.softdtw_variants.cdist_soft_dtw and not tslearn.metrics.dtw_variants.dtw or tslearn.metrics.dtw_variants.dtw_'
     displayName: 'Test'
 
 
@@ -124,8 +121,7 @@ jobs:
       python -m pip install coverage pytest-cov
       python -m pip install cesium pandas stumpy
       python -m pip install scikit-learn==1.2
-      python -m pip install tensorflow==2.9.0
-      python -m pytest -v tslearn/ --doctest-modules --cov=tslearn
+      export KERAS_BACKEND="torch" && python -m pytest -v tslearn/ --doctest-modules --cov=tslearn
     displayName: 'Test'
 
   # Upload coverage to codecov.io
@@ -139,8 +135,8 @@ jobs:
     vmImage: 'macOS-12'
   strategy:
     matrix:
-      Python38:
-        python.version: '3.8'
+      # Python38:
+      #   python.version: '3.8'
       Python39:
         python.version: '3.9'
       Python310:
@@ -176,8 +172,7 @@ jobs:
       set -xe
       python -m pip install pytest pytest-azurepipelines
       python -m pip install scikit-learn==1.2
-      python -m pip install tensorflow==2.9.0
-      python -m pytest -v tslearn/ --doctest-modules -k 'not test_all_estimators'
+      export KERAS_BACKEND="torch" && python -m pytest -v tslearn/ --doctest-modules -k 'not test_all_estimators'
     displayName: 'Test'
 
 
@@ -186,9 +181,9 @@ jobs:
     vmImage: 'windows-latest'
   strategy:
     matrix:
-      Python38:
-        python_ver: '38'
-        python.version: '3.8'
+      # Python38:
+      #   python_ver: '38'
+      #   python.version: '3.8'
       Python39:
         python_ver: '39'
         python.version: '3.9'
@@ -219,6 +214,5 @@ jobs:
   - script: |
       python -m pip install pytest pytest-azurepipelines
       python -m pip install scikit-learn==1.2
-      python -m pip install tensorflow==2.9.0
-      python -m pytest -v tslearn/ --doctest-modules --ignore tslearn/tests/test_estimators.py --ignore tslearn/utils/cast.py
+      set KERAS_BACKEND=torch && python -m pytest -v tslearn/ --doctest-modules --ignore tslearn/tests/test_estimators.py --ignore tslearn/utils/cast.py
     displayName: 'Test'
diff --git a/docs/examples/classification/plot_shapelet_distances.py b/docs/examples/classification/plot_shapelet_distances.py
index fdd9a3073..c11995564 100644
--- a/docs/examples/classification/plot_shapelet_distances.py
+++ b/docs/examples/classification/plot_shapelet_distances.py
@@ -24,7 +24,7 @@
 from tslearn.datasets import CachedDatasets
 from tslearn.preprocessing import TimeSeriesScalerMinMax
 from tslearn.shapelets import LearningShapelets
-from tensorflow.keras.optimizers import Adam
+from keras.optimizers import Adam
 
 # Set a seed to ensure determinism
 numpy.random.seed(42)
@@ -45,7 +45,7 @@
 # Define the model and fit it using the training data
 shp_clf = LearningShapelets(n_shapelets_per_size=shapelet_sizes,
                             weight_regularizer=0.0001,
-                            optimizer=Adam(lr=0.01),
+                            optimizer=Adam(learning_rate=0.01),
                             max_iter=300,
                             verbose=0,
                             scale=False,
diff --git a/docs/examples/classification/plot_shapelet_locations.py b/docs/examples/classification/plot_shapelet_locations.py
index c55987d27..556af2251 100644
--- a/docs/examples/classification/plot_shapelet_locations.py
+++ b/docs/examples/classification/plot_shapelet_locations.py
@@ -25,7 +25,7 @@
 from tslearn.preprocessing import TimeSeriesScalerMinMax
 from tslearn.shapelets import LearningShapelets, \
     grabocka_params_to_shapelet_size_dict
-from tensorflow.keras.optimizers import Adam
+from keras.optimizers import Adam
 
 # Set a seed to ensure determinism
 numpy.random.seed(42)
@@ -51,7 +51,7 @@
 # Define the model and fit it using the training data
 shp_clf = LearningShapelets(n_shapelets_per_size=shapelet_sizes,
                             weight_regularizer=0.001,
-                            optimizer=Adam(lr=0.01),
+                            optimizer=Adam(learning_rate=0.01),
                             max_iter=250,
                             verbose=0,
                             scale=False,
diff --git a/docs/examples/classification/plot_shapelets.py b/docs/examples/classification/plot_shapelets.py
index 150b70a19..202b5ebad 100644
--- a/docs/examples/classification/plot_shapelets.py
+++ b/docs/examples/classification/plot_shapelets.py
@@ -16,7 +16,7 @@
 
 import numpy
 from sklearn.metrics import accuracy_score
-import tensorflow as tf
+import keras
 import matplotlib.pyplot as plt
 
 from tslearn.datasets import CachedDatasets
@@ -49,7 +49,7 @@
 # Define the model using parameters provided by the authors (except that we
 # use fewer iterations here)
 shp_clf = LearningShapelets(n_shapelets_per_size=shapelet_sizes,
-                            optimizer=tf.optimizers.Adam(.01),
+                            optimizer=keras.optimizers.Adam(learning_rate=.01),
                             batch_size=16,
                             weight_regularizer=.01,
                             max_iter=200,
diff --git a/docs/requirements_rtd.txt b/docs/requirements_rtd.txt
index 6859395e1..b1ab951cb 100644
--- a/docs/requirements_rtd.txt
+++ b/docs/requirements_rtd.txt
@@ -8,7 +8,7 @@ ipykernel
 nbsphinx
 sphinx-gallery
 pillow
-tensorflow>=2
+keras>=3
 Pygments
 numba
 sphinx_bootstrap_theme
diff --git a/requirements.txt b/requirements.txt
index b23ded6e5..79ce5ae66 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,7 +3,7 @@ numba
 scipy
 scikit-learn
 joblib>=0.12
-tensorflow>=2
+keras>=3
 pandas
 cesium
 h5py
diff --git a/requirements_nocast.txt b/requirements_nocast.txt
index 270f87f5d..2d71623df 100644
--- a/requirements_nocast.txt
+++ b/requirements_nocast.txt
@@ -3,5 +3,5 @@ numba
 scipy
 scikit-learn
 joblib>=0.12
-tensorflow>=2
+keras>=3
 h5py
diff --git a/tslearn/shapelets/__init__.py b/tslearn/shapelets/__init__.py
index cb5bc5707..b84c11e17 100644
--- a/tslearn/shapelets/__init__.py
+++ b/tslearn/shapelets/__init__.py
@@ -1,7 +1,7 @@
 """
 The :mod:`tslearn.shapelets` module gathers Shapelet-based algorithms.
 
-It depends on the `tensorflow` library for optimization (TF2 is required).
+It depends on the `keras` library (Keras3+ is required).
 
 **User guide:** See the :ref:`Shapelets <shapelets>` section for further 
 details.
diff --git a/tslearn/shapelets/shapelets.py b/tslearn/shapelets/shapelets.py
index 50476a2ba..6b000ffac 100644
--- a/tslearn/shapelets/shapelets.py
+++ b/tslearn/shapelets/shapelets.py
@@ -1,19 +1,20 @@
-from tensorflow.keras.models import Model, model_from_json
-from tensorflow.keras.layers import (InputSpec, Dense, Conv1D, Layer, Input,
-                                     concatenate, add)
-from tensorflow.keras.metrics import (categorical_accuracy,
-                                      categorical_crossentropy,
-                                      binary_accuracy, binary_crossentropy)
-from tensorflow.keras.utils import to_categorical
+import keras
+from keras.models import Model, model_from_json
+from keras.layers import (InputSpec, Dense, Conv1D, Layer, Input,
+                                    concatenate, add)
+from keras.metrics import (categorical_accuracy,
+                                    categorical_crossentropy,
+                                    binary_accuracy, binary_crossentropy)
+from keras.utils import to_categorical
+from keras.regularizers import l2
+from keras.initializers import Initializer
+import keras.ops as ops
+
 from sklearn.base import ClassifierMixin, TransformerMixin
 from sklearn.utils import check_array, check_X_y
 from sklearn.utils.validation import check_is_fitted
 from sklearn.utils.multiclass import unique_labels
-from tensorflow.keras.regularizers import l2
-from tensorflow.keras.initializers import Initializer
-import tensorflow.keras.backend as K
 import numpy
-import tensorflow as tf
 
 import warnings
 
@@ -35,33 +36,46 @@ class GlobalMinPooling1D(Layer):
         
     Examples
     --------
-    >>> x = tf.constant([5.0, numpy.nan, 6.8, numpy.nan, numpy.inf])
-    >>> x = tf.reshape(x, [1, 5, 1])
+    >>> x = numpy.array([5.0, numpy.nan, 6.8, numpy.nan, numpy.inf])
+    >>> x = x.reshape([1, 5, 1])
     >>> GlobalMinPooling1D()(x).numpy()
     array([[5.]], dtype=float32)
     """
 
-    def __init__(self, **kwargs):
+    def __init__(self, data_format=None, keepdims=False, **kwargs):
         super().__init__(**kwargs)
+
+        self.data_format = (
+            "channels_last" if data_format is None else data_format
+        )
+        self.keepdims = keepdims
         self.input_spec = InputSpec(ndim=3)
 
     def compute_output_shape(self, input_shape):
         return input_shape[0], input_shape[2]
 
-    def call(self, inputs, **kwargs):
-        inputs_without_nans = tf.where(tf.math.is_finite(inputs),
-                                       inputs,
-                                       tf.zeros_like(inputs) + numpy.inf)
-        return tf.reduce_min(inputs_without_nans, axis=1)
+    def call(self, inputs):
+        steps_axis = 1 if self.data_format == "channels_last" else 2
+        inputs_without_nans = ops.where(ops.isfinite(inputs), 
+                                        inputs, 
+                                        ops.zeros_like(inputs) + ops.max(inputs[ops.isfinite(inputs)]))
+        return ops.min(inputs_without_nans, axis=steps_axis, keepdims=self.keepdims)
 
 
 class GlobalArgminPooling1D(Layer):
-    """Global min pooling operation for temporal data.
+    """Global argmin pooling operation for temporal data.
     # Input shape
         3D tensor with shape: `(batch_size, steps, features)`.
     # Output shape
         2D tensor with shape:
         `(batch_size, features)`
+        
+    Examples
+    --------
+    >>> x = numpy.array([5.0, numpy.nan, 6.8, numpy.nan, numpy.inf])
+    >>> x = x.reshape([1, 5, 1])
+    >>> GlobalArgminPooling1D()(x).numpy()
+    array([[0.]], dtype=float32)
     """
 
     def __init__(self, **kwargs):
@@ -72,7 +86,10 @@ def compute_output_shape(self, input_shape):
         return input_shape[0], input_shape[2]
 
     def call(self, inputs, **kwargs):
-        return K.cast(K.argmin(inputs, axis=1), dtype=K.floatx())
+        inputs_without_nans = ops.where(ops.isfinite(inputs), 
+                                        inputs, 
+                                        ops.zeros_like(inputs) + ops.max(inputs[ops.isfinite(inputs)]))
+        return ops.cast(ops.argmin(inputs_without_nans, axis=1), dtype=float)
 
 
 def _kmeans_init_shapelets(X, n_shapelets, shp_len, n_draw=10000):
@@ -106,7 +123,7 @@ def __call__(self, shape, dtype=None):
         shapelets = _kmeans_init_shapelets(self.X_,
                                            n_shapelets,
                                            shp_len)[:, :, 0]
-        return tf.convert_to_tensor(shapelets, dtype=K.floatx())
+        return ops.convert_to_tensor(shapelets, dtype=float)
 
     def get_config(self):
         return {'data': self.X_}
@@ -140,11 +157,11 @@ def build(self, input_shape):
 
     def call(self, x, **kwargs):
         # (x - y)^2 = x^2 + y^2 - 2 * x * y
-        x_sq = K.expand_dims(K.sum(x ** 2, axis=2), axis=-1)
-        y_sq = K.reshape(K.sum(self.kernel ** 2, axis=1),
+        x_sq = ops.expand_dims(ops.sum(x ** 2, axis=2), axis=-1)
+        y_sq = ops.reshape(ops.sum(self.kernel ** 2, axis=1),
                          (1, 1, self.n_shapelets))
-        xy = K.dot(x, K.transpose(self.kernel))
-        return (x_sq + y_sq - 2 * xy) / K.int_shape(self.kernel)[1]
+        xy = ops.dot(x, ops.transpose(self.kernel))
+        return (x_sq + y_sq - 2 * xy) / ops.shape(self.kernel)[1]
 
     def compute_output_shape(self, input_shape):
         return input_shape[0], input_shape[1], self.n_shapelets
@@ -426,7 +443,7 @@ def fit(self, X, y):
         self._check_series_length(X)
 
         if self.random_state is not None:
-            tf.keras.utils.set_random_seed(seed=self.random_state)
+            keras.utils.set_random_seed(seed=self.random_state)
 
         n_ts, sz, d = X.shape
         self._X_fit_dims = X.shape
@@ -567,7 +584,7 @@ def locate(self, X):
         >>> X[0, 4:7, 0] = numpy.array([1, 2, 3])
         >>> y = [1, 0, 0]
         >>> # Data is all zeros except a motif 1-2-3 in the first time series
-        >>> clf = LearningShapelets(n_shapelets_per_size={3: 1}, max_iter=0,
+        >>> clf = LearningShapelets(n_shapelets_per_size={3: 1}, max_iter=1,
         ...                     verbose=0)
         >>> _ = clf.fit(X, y)
         >>> weights_shapelet = [
@@ -780,7 +797,7 @@ def get_weights(self, layer_name=None):
         --------
         >>> from tslearn.generators import random_walk_blobs
         >>> X, y = random_walk_blobs(n_ts_per_blob=100, sz=256, d=1, n_blobs=3)
-        >>> clf = LearningShapelets(n_shapelets_per_size={10: 5}, max_iter=0,
+        >>> clf = LearningShapelets(n_shapelets_per_size={10: 5}, max_iter=1,
         ...                     verbose=0)
         >>> clf.fit(X, y).get_weights("classification")[0].shape
         (5, 3)
@@ -816,7 +833,7 @@ def set_weights(self, weights, layer_name=None):
         --------
         >>> from tslearn.generators import random_walk_blobs
         >>> X, y = random_walk_blobs(n_ts_per_blob=10, sz=16, d=1, n_blobs=3)
-        >>> clf = LearningShapelets(n_shapelets_per_size={3: 1}, max_iter=0,
+        >>> clf = LearningShapelets(n_shapelets_per_size={3: 1}, max_iter=1,
         ...                     verbose=0)
         >>> _ = clf.fit(X, y)
         >>> weights_shapelet = [
diff --git a/tslearn/tests/test_estimators.py b/tslearn/tests/test_estimators.py
index 08108e38d..0cb62edd1 100644
--- a/tslearn/tests/test_estimators.py
+++ b/tslearn/tests/test_estimators.py
@@ -77,7 +77,7 @@ def _get_all_classes():
                 # keras is likely not installed
                 warnings.warn('Skipped common tests for shapelets '
                               'as it could not be imported. keras '
-                              '(and tensorflow) are probably not '
+                              'is probably not '
                               'installed!')
                 continue
             elif name.endswith('pytorch_backend'):
diff --git a/tslearn/tests/test_shapelets.py b/tslearn/tests/test_shapelets.py
index f58a519b9..466f10b5c 100644
--- a/tslearn/tests/test_shapelets.py
+++ b/tslearn/tests/test_shapelets.py
@@ -8,9 +8,9 @@
 
 
 def test_shapelets():
-    pytest.importorskip('tensorflow')
+    pytest.importorskip('keras')
     from tslearn.shapelets import LearningShapelets
-    import tensorflow as tf
+    import keras
 
     n, sz, d = 15, 10, 2
     rng = np.random.RandomState(0)
@@ -27,7 +27,7 @@ def test_shapelets():
     clf = LearningShapelets(n_shapelets_per_size={2: 5},
                             max_iter=1,
                             verbose=0,
-                            optimizer=tf.optimizers.Adam(.1),
+                            optimizer=keras.optimizers.Adam(learning_rate=.1),
                             random_state=0)
     cross_validate(clf, time_series, y, cv=2)
 
@@ -62,7 +62,7 @@ def test_shapelets():
 
 
 def test_shapelet_lengths():
-    pytest.importorskip('tensorflow')
+    pytest.importorskip('keras')
     from tslearn.shapelets import LearningShapelets
 
     # Test variable-length
@@ -97,7 +97,7 @@ def test_shapelet_lengths():
 
 
 def test_series_lengths():
-    pytest.importorskip('tensorflow')
+    pytest.importorskip('keras')
     from tslearn.shapelets import LearningShapelets
 
     # Test long shapelets