The paper presents a family of kernel mean shrinkage estimators. These estimators generalize the ones proposed in \cite{journals/jmlr/FukumizuSG13} and can incoporate useful domain knowledge through spetral filters. Here is a summary of interesting contributions:
1. Theorem 1 that shows the consistency and admissibility of kmse presented in \cite{journals/jmlr/FukumizuSG13}.
2. The idea of spectral kmse (its use in this unsupervised setting) and similarity of final form with the supervised setting.
3. Theorem 5 that shows consistency of the proposed spectral kmse.

This paper proposes a method to obtain a non-vacuous bound on generalization error by optimizing the PAC-Bayes bound directly. The interesting part is that the authors leverage the black magic of neural net itself to bound the neural net. In order to find the optimal Q, the authors' loss function is an empirical err term plus the $KL(Q|P)$, where they choose the prior $P$ to be $N(0, \lambda I)$, and they also provide justification for choosing the right $\lambda$. Overally, this objective is similar to the variational inference in the Bayesian neural net, and the author is able to obtain a test error bound of $17\%$ on MNIST, while the tradition bounds will be mostly meaningless.

Salman et al. combined randomized smoothing with adversarial training based on an attack specifically designed against smoothed classifiers. Specifically, they consider the formulation of randomized smoothing by Cohen et al. [1]; here, Gaussian noise around the input (adversarial or clean) is sampled and the classifier takes a simple majority vote. In [1], Cohen et al. show that this results in good bounds on robustness. In this paper, Salman et al. propose an adaptive attack against randomized smoothing. Essentially, they use a simple PGD attack to attack a smoothed classifier, i.e., maximize the cross entropy loss of the smoothed classifier. To make the objective tractable, Monte Carlo samples are used in each iteration of the PGD optimization. Based on this attack, they do adversarial training, with adversarial examples computed against the smoothed (and adversarially trained) classifier. In experiments, this approach outperforms the certified robustness by Cohen et al. on several datasets.

[1] Jeremy M. Cohen, Elan Rosenfeld and J. Zico Kolter. Certified Adversarial Robustness via Randomized Smoothing. ArXiv, 1902.02918, 2019.

Also find this summary at [davidstutz.de](https://davidstutz.de/category/reading/).

####  Mask R-CNN framework for instance segmentation
### Goal:
* classify individual objects
*  localize each using a bounding box, 
* semantic segmentation
https://i.imgur.com/XfBRa5O.png
*  classify each pixel into a fixed set of categories without differentiating object instances.
* extends Faster R-CNN  by adding a branch for predicting segmentation masks on each Region of Interest (RoI), in parallel with the existing branch for classification and bounding box regression.
*  FCN applied to each RoI, predicting a segmentation mask in a pixel-to-pixel manner
1. RoIAlign:
 * Used to fix the misalignment that faithfully preserves exact spatial locations
 * improves mask accuracy by relative 10% to 50%, fast speed
2. Decouple mask and class prediction:
 * predict a binary mask for each class independently, without competition among classes

 History:
* RCNN: The Region-based CNN (R-CNN) approach to bounding-box object detection
* Fast RCNN:  Speeding up and Simplifying R-CNN
 *  RoI (Region of Interest) Pooling
 *  jointly train the CNN, classifier, and bounding box regressor in a single model
*  Faster R-CNN - Speeding Up Region Proposal
 * reuse the same CNN results for region proposals instead of running a separate selective search algorithm it can be done by Region Proposal Network
 * only one CNN needs to be trained 

 Related Work
*  Instance Segmentation: “fully convolutional instance segmentation” (FCIS)
* Faster R-CNN: * Region Proposal Network (RPN), proposes candidate object bounding boxes
* Fast R-CNN [12], extracts features using RoIPool from each candidate box and performs classification and bounding-box regression
*  Mask R-CNN: Mask R-CNN adopts the same two-stage of Faster RCNN And has third stage i.e binary mask for each RoI
* Mask Representation: pixel to pixel representation of image done by RoIAlign layer (7X7)
#### Network Architecture
*  convolutional backbone architecture used for feature extraction over an entire image (ResNet-50-C4, FPN)
*  network head for bounding-box recognition (classification and regression) and mask prediction
https://i.imgur.com/pUvKdmx.png
#### Training:
* Images resized:800 pixel                                     
* mini-batch : 2 images per GPU
* N : 64
* train: on 8 GPUs for 160k iterations
* learning : 0.02
*  train images: 80K
* val images: 35K
* minival:5K
https://i.imgur.com/6ZLpewi.png
https://i.imgur.com/5o3um0Y.png

https://www.youtube.com/watch?v=e-WB4lfg30M


This technique is a combination of two powerful machine learning algorithms:
- convolutional neural networks are excellent at image classification, i.e., finding out what is seen on an input image,
- recurrent neural networks that are capable of processing a sequence of inputs and outputs, therefore it can create sentences of what is seen on the image.

Combining these two techniques makes it possible for a computer to describe in a sentence what is seen on an input image.