Tue Nov 01 2022
Mon Oct 31 2022

DiffusER: Discrete Diffusion via Edit-based Reconstruction

text generation
natural language processing
generative models
improve text generation
allow for revision of existing text
condition generation on a prototype and continue revising based on previous edit steps

DiffusER rivals autoregressive models on several tasks spanning machine translation, summarization, and style transfer.

DiffusER is a strong generative model in general, rivaling autoregressive models on several tasks spanning machine translation, summarization, and style transfer. It can also perform other varieties of generation that standard autoregressive models are not well-suited for.

Lila: A Unified Benchmark for Mathematical Reasoning

mathematical reasoning assessment
artificial intelligence
mathematical reasoning
improve AI systems in mathematical reasoning
measure out-of-distribution performance and robustness to language perturbation
engaging in multi-tasking to improve performance

Proposes Lila, a unified mathematical reasoning benchmark consisting of 23 diverse tasks along four dimensions.

Lila offers a unified benchmark for evaluating and improving AI systems in mathematical reasoning consisting of 23 diverse tasks along four dimensions

Preventing Verbatim Memorization in Language Models Gives a False Sense of Privacy

neural language models
natural language processing
data memorization
understanding the risks associated with models regurgitating training data
evaluation of potential countermeasures

Argues that verbatim memorization definitions are too restrictive and fail to capture more subtle forms of memorization.

Defining memorization is a difficult and crucial open question for neural language models.

Sun Oct 30 2022
Thu Oct 27 2022
Wed Oct 26 2022
Tue Oct 25 2022