Update README.md

README.md

@@ -258,3 +258,7 @@ On COCO and Flickr30k retrieval, we report zero-shot accuracy, on image tasks, w
 ## Limitations
 
 Currently, FLAVA has many limitations. The image classification accuracy is not on par with CLIP on some of the tasks while text accuracy is not on par with BERT on some of the tasks suggesting possible room for improvement. FLAVA also doesn't work well on tasks containing scene text given the lack of scene text in most public datasets. Additionally, similar to CLIP, our approach to testing FLAVA also has an important limitation in the case of image tasks, where we use linear probes to evaluate FLAVA and there is evidence suggesting that linear probes can underestimate model performance.
+
+## Feedback/Questions
+
+Please email Amanpreet at `amanpreet [at] nyu [dot] edu` for questions.