Remaining Challenges in Deep Learning Based NLP

Explore the remaining challenges in Deep Learning-based Natural Language Processing (NLP) in this insightful 31-minute conference talk. Delve into the limitations of current neural models, including their reliance on large training datasets, potential biases in performance metrics, and lack of robustness. Examine the shortcomings of distributional word embeddings and sentence-level representations. Investigate the implications of limited interpretability in neural networks, affecting both debugging processes and fairness issues. Gain a critical perspective on the state of AI in language processing and understand the areas that still require significant improvement in the field of NLP.

Introduction
About me
Popular media examples
Adding more layers
Representation learning
Recurrent neural networks
Big data problem
Overfitting
Life cycle
Distributional similarity
Sentence embedding
Not robust enough
Lack of interpretability
Questions