Deep Learning - D.I.Y.
summary
Hands-on repo and VirtualBox appliance contains Fedora 25, Jupyter, pretrained ImageNet models (Inception v3 / ResNet) and runnable TensorFlow examples; demos include the TensorFlow playground and a spectrogram-to-image trick that turns spoken digits into CNN inputs. A CNN trained on locally recorded digit WAVs runs on the VM and is trained in under three minutes, reaching perfect validation on a tiny held-out set (about 20 samples) and using an SVM on error fingerprints to reduce mistakes.
AI-written summary · about
chapters (10)
- 0:00Start
- 3:53WHAT CAN BE DONE NOW
- 4:11SPEECH RECOGNITION
- 7:37MULTI-LAYER
- 16:17IMAGE CLASSIFICATION
- 17:31WHAT IS A CNN?
- 18:56CNN FILTER
- 20:45CNN FLOW
- 54:22DEEP LEARNING MEETUP GROUP
- 55:3110-WEEK DEEP LEARNING DEVELOPER COURSE
related talks
01h 34mDeep Learning Hands-on Workshop
01h 06mDeep Learning with TensorFlow
26mDeep Learning in OpenCV
01h 39mGoogle Colaboratory and Python: Deep Learning basics for CV tasks
01h 07mDeep Learning with TensorFlow
57mDeep Learning with C++
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
01h 04mWhat you always wanted to know about Deep Learning, but were afraid to ask