Visual Storytelling
Deep Compression, DSD Training and EIE: Deep Neural Network Model Compression, Regularization and Hardware Acceleration
Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on mobile phones and embedded systems with limited hardware resources. To address this limitation, this talk first introduces “Deep Compression” that…
From Models of Language Understanding to Agents of Language Use
Neural network-style algorithms and architectures have recently been applied to achieve notable improvements to various language applications. However, when it comes to general language understanding, the performance of these models is still far from human-like.…
Skrybe: Designing Features to Improve Real-Time Captioning for Deaf & Hard of Hearing Students
Today, real-time captioning services enable deaf and hard of hearing (DHH) students to participate in classes with hearing students and teachers. However, these services were not designed specifically with DHH students in mind. Current solutions…