update documentation

README.md

@@ -23,7 +23,7 @@ The benefits of a technique known as *self-attention* are explored in **aa_atten
 an otherwise similar model except with a self-attention layer.
 ![Self-Attention Generative Adversarial Networks (SAGAN)](https://arxiv.org/abs/1805.08318)
 explains how to use self-attention in a GAN environment. This model adapts the layer to Adversarial Autoencoder.
-Additionally this model uses adversarial loss in the reconstructed image, resulting in more "accurate" and "humanlike" features.
+Additionally this model uses adversarial loss for the reconstructed image, resulting in more "accurate" and "humanlike" features.
 
 
 ![The latent space with self-attention and color](attention_comparison.mp4)