Most ARM chips can't run decent AI models. Introducing EfficientSpeech, a 266k-param TTS model. Low cost ARM chips like in RPi4 can generate 104sec of speech mel spec in 1sec. Here's an AI-generated video w/ voice from EfficientSpeech. Info: https://t.co/z0ZzZHZEMH #ICASSP2023
@arXiv_Daily Simple yet effective idea: Remove inefficient top-most layers & replace them with an efficient head. For VWW, param count reduced by 93% with only 0.65% accuracy decrease. Counterintuitively, the quantized pruned net increased its accuracy on ARM Cortex M0.
"The UP National Engineering Center Analytics and Data Science Certifications announced the development on Wednesday." https://t.co/UQ7QXEe2v9 via @gmanews
Idea: If data augmentation improves model generalization, why not use it to generate 2 new inputs and force the representations to agree. Result: Additional model performance improvement. Comparison: Unlike Label Smoothing, the performance of our method, AgMax, is consistent.
Improving Model Generalization by Agreement of Learned Representations from Data Augmentation
https://t.co/QuQDB9MWzW
by @jacobe#ComputerVision#ImageNet
I am excited to share my latest work: 8-bit optimizers – a replacement for regular optimizers. Faster 🚀, 75% less memory 🪶, same performance📈, no hyperparam tuning needed 🔢. 🧵/n
Paper: https://t.co/V5tjOmaWvD
Library: https://t.co/JAvUk9hrmM
Video: https://t.co/TWCNpCtCap
We’re introducing GSLM, the first language model that breaks free completely of the dependence on text for training. This “textless NLP” approach learns to generate expressive speech using only raw audio recordings as input. Learn more and get the code:
https://t.co/kRkUaFyZWb
@arXiv_Daily Data Augmentation for STR will be presented at ICCV 2021 Workshop on Interactive Labeling and Data Augmentation for Vision. GitHub: https://t.co/OLfGeLQRfc
@arXiv_Daily It took us more than a year building, collecting, annotating, validating and benchmarking this dataset.
Dataset: https://t.co/t13w9wWlBo
To appear at #CVPR2021 Workshop: https://t.co/H84qAjdSjE
Yesterday, my former grad student Daryl gave a talk at Sony CSL Paris about his thesis on Next View Policy for 3D Reconstruction. Youtube: https://t.co/Fxu4B8e6eU