Methods for learning from noisy data, using phonetic embeddings to improve entity resolution, and quantization-aware training are a few of the highlights.