mish activation

keras/activations.py

@@ -35,6 +35,15 @@ def softmax(x, axis=-1):
         raise ValueError('Cannot apply softmax to a tensor that is 1D. '
                          'Received input: %s' % x)
 
+def mish(x):
+    """Mish: A Self Regularized Non-Monotonic Neural Activation Function (https://arxiv.org/abs/1908.08681v2)
+    Overperforms than both ReLU and Swish.
+    # Arguments
+        x: Input tensor.
+    # Returns
+        x * tanh(softplus(x))
+    """
+    return x*K.tanh(K.softplus(x))
 
 def elu(x, alpha=1.0):
     """Exponential linear unit.