Submitted a PR for adding a new activation Sparsemax activation function in keras implementing it in jax, numpy, tensorflow and pytorch as keras supports all these backends.

Submitted a PR for fixing a Sparse Categorical Crossentropy loss and Mixed precision while using float16 and bfloat16 data type and overflow caused due to this.

Submitted a PR for creating a new parameter, initial_value_threshold in ModelCheckpoint Callback that only overwrites the model weights already saved if the performance of current model is better than this value.

Submitted a PR for implementing the different scoring methods in Luong-Style Attention. Thus creating a new parameter for Attention Layer in Keras called score_mode.

Submitted a PR for implementing Thresholded_Relu with different data types, essentially making it compatible with list datatype.

Submitted a PR for raising a warning if 2 labels are only present in categorical crossentropy loss rather than binary loss.

Implementation of the paper MLP-Mixer: An all-MLP Architecture for Vision, MLP-mixer in Tensorflow/Keras.