Seminars
Abstract
Model robustness and spurious correlations have received increasing attention in the NLP community, both in methods and evaluation. The term “spurious correlation” is overloaded though and can refer to any undesirable shortcuts learned by the model, as judged by domain experts.
When designing mitigation algorithms, we often (implicitly) assume that a spurious feature is irrelevant for prediction. However, many features in NLP (e.g. word overlap and negation) are not spurious in the sense that the background is spurious for classifying objects in an image. In contrast, they carry important information that’s needed to make predictions by humans. In this talk, we argue that it is more productive to characterize features in terms of their necessity and sufficiency for prediction. We then discuss the implications of this categorization in representation, learning, and evaluation.
Biography
He He is an Assistant Professor in the Department of Computer Science and the Center for Data Science at New York University. She obtained her PhD in Computer Science at the University of Maryland, College Park. Before joining NYU, she spent a year at AWS AI and was a post-doc at Stanford University before that. She is interested in building robust and trustworthy NLP systems in human-centered settings. Her recent research focus includes robust language understanding, collaborative text generation, and understanding capabilities and issues of large language models.
Abstract
Transformers are essential to pretraining. As we approach 5 years of BERT, the connection between attention as architecture and transfer learning remains key to this central thread in NLP. Other architectures such as CNNs and RNNs have been used to replicate pretraining results, but these either fail to reach the same accuracy or require supplemental attention layers. This work revisits the semanal BERT result and considers pretraining without attention. We consider replacing self-attention layers with recently developed approach for long-range sequence modeling and transformer architecture variants. Specifically, inspired by recent papers like the structured space space sequence model (S4), we use simple routing layers based on state-space models (SSM) and a bidirectional model architecture based on multiplicative gating. We discuss the results of the proposed Bidirectional Gated SSM (BiGS) and present a range of analysis into its properties. Results show that architecture does seem to have a notable impact on downstream performance and a different inductive bias that is worth exploring further.
Biography
Abstract
Recent advances in large pretrained language models have unlocked new exciting applications for Natural Language Generation for creative tasks, such as lyrics or humour generation. In this talk we will discuss recent works by our team at Alexa AI and discuss current challenges: (1) Pun understanding and generation: We release new datasets for pun understanding and the novel task of context-situated pun generation, and demonstrate the value of our annotations for pun classification and generation tasks. (2) Song lyric generation: we design a hierarchical lyric generation framework that enables us to generate pleasantly-singable lyrics without training on melody-lyric aligned data, and show that our approach is competitive with strong baselines supervised on parallel data. (3) Create with Alexa: a multimodal story creation experience recently launched on Alexa devices, which leverages story text generation models in tandem with story visualization and background music generation models to produce multimodal stories for kids.
Biography
Alessandra Cervone is an Applied Scientist in the Natural Understanding team at Amazon Alexa AI. Alessandra holds an MSc in Speech and Language Processing from University of Edinburgh and a PhD in CS from University of Trento (Italy). During her PhD, Alessandra worked on computational models of coherence in open-domain dialogue advised by Giuseppe Riccardi. In the first year of the PhD, she was the team leader of one of the teams selected to compete in the first edition of the Alexa Prize. More recently, her research interests have been focused on natural language generation and its evaluation, in particular in the context of creative AI applications.
