Warning
This page is located in archive.





Lectures

date week topic slides + videos
19.09.2022 1 Lec 1: Machine learning 101: model, loss, learning, issues, regression, classification 01_intro_learning_regression_classification.pdf
lec1_recording
26.09.2022 2 Lec 2: Under the hood of a linear classifier: two-class and multi-class linear classifier on RGB images 02_classification.pdf
lec2_recording
03.10.2022 3 Lec 3: Where the hell does the loss come from? MAP and ML estimate, KL divergence and losses. 03_mle.pdf
lec3_recording
10.10.2022 4 Lec 4: The story of the cat's brain surgery: fully-connected NN + fast backpropagation via Vector-Jacobian-Product (VJP), cortex + convolutional layer
04b_convnets.pdf
lec4_recording
17.10.2022 5 Lec 5: Under the hood of auto-differentiation: Vector-Jacobian-Product (VJP) vs chainrule and multiplication of Jacobians, convolutional layer and its VJP 04a_fully_connected_neural_nets.pdf
lec5_recording
24.10.2022 6 Midterm test vir_2022_training_questions_midterm_test.pdf
(“pure SGD” = “SGD”, i.e. momentum = 0)
31.10.2022 7 Lec 6: Why is learning prone to fail? - Structural issues: layers + issues, batch-norm, drop-out layers.pdf
lec6_recording
07.11.2022 8 Lec 7: Why is learning prone to fail? - Optim. issues: optimization vs learning, KL divergence, SGD, momentum, convergence rate, Adagrad, RMSProp, AdamOptimizer, diminishing/exploding gradient, oscillation, double descent training.pdf
lec7_recording
14.11.2022 9 Lec 8: What can('t) we do with a deep net?: Classification (ResNet, Squeeze and Excitation Nets), Segmentation (DeepLab), Detection (Yolo, fast-RCNN), Regression (OpenPose), Spatial Transformer Nets, Memory and attention (recurrent nets, Image transformers with attention module) 08a_architectures_class_segm_simplified.pdf
08b_architectures_simple_reg_det_stn.pdf
lec8_recording
21.11.2022 10 Lec 9: Reinforcement learning: Approximated Q-learning, DQN, DDPG, Derivation of the policy gradient (REINFORCE), A2C, Inverse RL, Applications,

Lec 9 3/4:Learning to optimize: Backpropagation through unconstrained and constrained optimization problems, application (end-to-end differentiable modules cvxpy, gradSLAM, gradMPC, gradODE, pytorch3d)

Lec 9 9/10:Self-supervision: (pseudo-labeling, privileged information, monodepth)
09_reinforcement_learning.pdf
lec9_recording
28.11.2022 11 Lec 10: Generative Adversial Networks: Guest lecture by David Coufal, ÚI AV ČR lec10_recording
vir2022_a.pdf
05.12.2022 12 Lec 11: Normalizing Flows: Guest lecture by David Coufal, ÚI AV ČR lec11_recording
vir2022_b.pdf
12.12.2022 13 Exam test vir_2022_training_questions_exam_test.pdf
09.01.2023 15 Lec 12: Deep learning for satellite imagery: Guest lecture of Michal Reinstein (chief scientist in Spaceknow)

The playlist of all lecture recordings is here.





Lectures 2021

date week topic slides + videos
20.09.2021 1 Lec 1: MLE regression: derivation of L2 loss, prior lec_01_MLE_regression
27.09.2021 2 Lec 2: MLE classifier: derivation of cross-entropy loss and logistic loss and linear classifier lec_02_MLE_classifiction(pdf)
lec_02_recording
codes
04.10.2021 3 Lec 3: Fully-connected network: computational graph + backpropagation lec_03_NN(pdf)
lec_03_recording
11.10.2021 4 Lec 4: ConvNet: convolutional layer + backpropagation lec_04_ConvNets(pdf)
lec_04_recording
18.10.2021 5 Lec 5: Training: SGD, momentum, convergence rate, Adagrad, RMSProp, AdamOptimizer, diminishing/exploding gradient, oscilation lec_05_training(pdf)
lec_05_recording
25.10.2021 6 Lec 6: Layers: Convolution, Activation functions, Batch-Instance Norm, MaxPooling, Losses + backpropagation
Test T1
lec_06_ConvNets
training_examples.zip
01.11.2021 7 Lec 7: Architectures I: classification (ResNet, Squeeze and `excitation Nets) lec_07_classification_segmentation(pdf)
lec_07_recording
08.11.2021 8 Lec 8: Architectures II: segmentation (DeepLab), pose regression (OpenPose), lec_08_classification_segmentation(pdf)
lec_08_recording
15.11.2021 9 Lec 9: Architectures III: detection (Yolo), depth regression, spatial transformer nets lec_09_regression_detection_stn(pdf)
SVTI streams and recordings
22.11.2021 10 Lec 10: Reinforcement Learning I: Approximated Q-learning, DQN, DDPG, Policy gradient (REINFORCE), lec_10_DQN_DDPG_REINFORCE(pdf)
lec_10_recording
29.11.2021 11 Lec 11: Generative Adversarial Networks:
guest lecture by David Coufal, ÚI AV ČR)
lect_11_GANs(pdf)
lec_11_recording
06.12.2021 12 Lec 12: Reinforcement Learning II: Derivation of the policy gradient, A2C, Inverse Reinforcement Learning, Applications, end-to-end differentiable modules (cvxpy, gradSLAM, pytorch3d) lec_12_A2C_IRL(pdf)
lec_12_recording
13.12.2021 13 Lec 13: Memory and attention: Recurrent nets, Image transformers with attention module, Depth completition (monodepth), Graph convolution, lec_13_memory_attention(pdf)
lec_13_recording
03.01.2022 15 Lec 14: Exam Test ET

All SVTI streams





Lectures 2020

date week topic slides+videos+codes
21.09.2020 1 Lec 1: Regression/Classification as MLE: derivation of L2-loss, cross-entropy loss, logistic loss

Lec 2: Neural networks: Fully-connected layer + computational graph + backpropagation
lec_1_video (incorrect audio sync!) vir_outline.pdfmle_01_regression.pdf mle_regression.py

lec2_internal_viewerlec_2_video mle_02_classification.pdf mle_linear_classifier.py
28.9.2020 2 State holidays
05.10.2020 3 Lec 3: ConvNet: convolutional layer + backpropagation lec3_internal_viewer lec3.video neural_nets.pdf, convnets.pdf convolution_codes.zip
12.10.2020 4 Lec 4: Training: SGD, momentum, convergence rate, Adagrad, RMSProp, AdamOptimizer, diminishing/exploding gradient, oscilation lec4_internal_viewer lec4_video
training.pdf (corrected slides)
19.10.2020 5 Lec 5: Layers: Convolution, Activation functions, Batch-Instance Norm, MaxPooling, Losses + backpropagation lec5_internal_viewer,
layers.pdf (corrected)
26.10.2020 6 Lec 6: Architectures I: classification (ResNet), segmentation (DeepLab)
Test T1
lec6_internal_viewer
architectures_i.pdf
02.11.2020 7 Lec 7: Learning from unlabelled data: Self-supervision: Contrastive learning, rotation, jigsaw, Colorization
Weak-supervision: Multiple-instance learning, physical constraints; guest lecture by Patrik Vacek
lec7_internal_viewer
unlabelled_data.pdf
09.11.2020 8 Lec 8: Architectures II: pose regression, detection (Yolo), depth regression, spatial transformer nets, LIFTs lec8_internal_viewer
architectures_ii.pdf
16.11.2020 9 Lec 9: Structured inputs: Recurrent neural networks, Convolution in 1,2 and 3D and other structures:
guest lecture by Teymur Azayev
lec9_internal_viewer
lecture_9_structure.pdf
23.11.2020 10 Lec 10: Reinforcement Learning I: DQN, GAE+TD(lambda), DDPG lec10_internal_viewer
reinforcement_learning.pdf
30.11.2020 11 Lec 11: Reinforcement Learning II: Policy gradients (REINFORCE), Inverse Reinforcement Learning, Applications,

What did not fit to lectures: domain transfer, MAML, monodepth, depth from symmetries, Pytorch3D, CvxPyLayer (it did not fit even here ;-))
lec11_internal_viewer
reinforcement_learning.pdf
07.12.2020 12 Lec 12: Generative Adversarial Networks:
guest lecture by David Coufal, ÚI AV ČR)
lec12_internal_viewer
gans_coufal_vir_2019
14.12.2020 13 Exam Test ET see BRUTE for results
04.01.2021 14 Selected teams present their semestral works (the rest presents during labs)
courses/b3b33vir/lectures/start.txt · Last modified: 2022/12/11 12:42 by zimmerk