[DL輪読会]Recent Advances in Autoencoder-Based Representation LearningDeep Learning JP
?
1. Recent advances in autoencoder-based representation learning include incorporating meta-priors to encourage disentanglement and using rate-distortion and rate-distortion-usefulness tradeoffs to balance compression and reconstruction.
2. Variational autoencoders introduce priors to disentangle latent factors, but recent work aggregates posteriors to directly encourage disentanglement.
3. The rate-distortion framework balances the rate of information transmission against reconstruction distortion, while rate-distortion-usefulness also considers downstream task usefulness.
1. This document discusses robots that utilize their own structure and morphology for locomotion and activity. It provides examples of robots that rely on physical dynamics and morphology for control rather than complex software or sensing.
2. Specific examples discussed include hopping robots that use the natural vibration of their structure for energy-efficient hopping, passive dynamic walkers that can walk solely through interaction with gravity and friction without actuation or control, and soft robots whose flexible materials and pneumatic networks allow intrinsically compliant motion.
3. The document argues that utilizing a robot's physical structure and materials for control can reduce the computational and sensing demands compared to systems relying solely on software control. This morphological computation is inspired by principles observed in biological systems
[DL輪読会]Recent Advances in Autoencoder-Based Representation LearningDeep Learning JP
?
1. Recent advances in autoencoder-based representation learning include incorporating meta-priors to encourage disentanglement and using rate-distortion and rate-distortion-usefulness tradeoffs to balance compression and reconstruction.
2. Variational autoencoders introduce priors to disentangle latent factors, but recent work aggregates posteriors to directly encourage disentanglement.
3. The rate-distortion framework balances the rate of information transmission against reconstruction distortion, while rate-distortion-usefulness also considers downstream task usefulness.
1. This document discusses robots that utilize their own structure and morphology for locomotion and activity. It provides examples of robots that rely on physical dynamics and morphology for control rather than complex software or sensing.
2. Specific examples discussed include hopping robots that use the natural vibration of their structure for energy-efficient hopping, passive dynamic walkers that can walk solely through interaction with gravity and friction without actuation or control, and soft robots whose flexible materials and pneumatic networks allow intrinsically compliant motion.
3. The document argues that utilizing a robot's physical structure and materials for control can reduce the computational and sensing demands compared to systems relying solely on software control. This morphological computation is inspired by principles observed in biological systems
1. ガウス過程回帰(GPR)の概要?導出と計算例
大阪大学 石黒研究室 博士後期課程2年 浦井健次
機械学習勉強会@大阪大学豊中キャンパス
参考文献
[1] 中村泰, 石黒浩: Gaussian process regression を用いた確率
的方策に対する方策勾配法, IEICE, 2012.
[2] 大羽成征, 石井信, 佐藤雅昭: ガウス過程法のオンライン学習,
IEICE, 2001.
[3] Carl Edward Rasmussen and Christopher K. Williams:
Gaussian Processes for Machine Learning. Massachusetts
Institute of Technology: MIT-Press, 2006.
[4] C.M. ビショップ, 元田, 栗田, 樋口, 松本, 村田: パターン認識と機
械学習(上)(下) ベイズ理論による統計的予測, Springer, 2007.
[5] Duy Nguyen-tuing and Jan Peters: Local gaussian
process regression for real time online model learning and
control, In In Advances in Neural Information Processing
Systems 22 (NIPS), 2008.
[6] Yuya Okadome, Kenji Urai, Yutaka Nakamura, Tetsuya
Yomo, and Hiroshi Ishiguro: Adaptive LSH based on the
particle swarm method with the attractor selection model
for fast approximation of Gaussian process regression,
Journal of Artificial Life and Robotics, 2014.