The paper proposes a method to perform joint instance and semantic segmentation. The method is fast as it is meant to run in an embedded environment (such as a robot). While the semantic map may seem redundant given the instance one, it is not as semantic segmentation is a key part of obtaining the instance map.

# Architecture

![image](https://user-images.githubusercontent.com/8659132/63187959-24cdb380-c02e-11e9-9121-77e0923e91c6.png)

The image is first put through a typical CNN encoder (specifically a ResNet derivative), followed by 3 separate decoders. The output of the decoder is at a low resolution for faster processing.

Decoders:
- Semantic segmentation: coupled with the encoder, it's U-Net-like. The output is a segmentation map.
- Instance center: for each pixel, outputs the confidence that it is the center of an object.
- Embedding: for each pixel, computes a 32 dimensional embedding. This embedding must have a low distance to embedding of other pixels of the same instance, and high distance to embedding of other pixels.

To obtain the instance map, the segmentation map is used to mask the other 2 decoder outputs to separate the embeddings and centers of each class. Centers are thresholded at 0.7, and centers with embedding distances lower than a set amount are discarded, as they are considered duplicates.

Then for each class, a similarity matrix is computed between all pixels from that class and centers from that class. Pixels are assigned to their closest centers, which represent different instances of the class.

Finally, the segmentation and instance maps are upsampled using the SLIC algorithm.

# Loss

There is one loss for each decoder head.
- Semantic segmentation: weighted cross-entropy
- Instance center: cross-entropy term modulated by a $\gamma$ parameter to counter the over-representation of the background over the target classes.
![image](https://user-images.githubusercontent.com/8659132/63286485-22659680-c286-11e9-9134-f1b823a34217.png)

- Embedding: composed of 3 parts, an attracting force between embeddings of the same instance, a repelling force between embeddings of different instances, and a l2 regularization on the embedding.
![image](https://user-images.githubusercontent.com/8659132/63286399-f1856180-c285-11e9-9136-feb6c4a555e5.png)
![image](https://user-images.githubusercontent.com/8659132/63286411-fcd88d00-c285-11e9-939f-0771579d8263.png)
$\hat{e}$ are the embeddings, $\delta_a$ is an hyper-parameter defining "close enough", and $\delta_b$ defines "far enough"

The whole model is trained jointly using a weighted sum of the 3 losses.

# Experiments and results

The authors test their method on the Cityscape dataset, which is composed of 5000 annotated images and 8 instance classes. They compare their methods both for semantic segmentation and instance segmentation.

![image](https://user-images.githubusercontent.com/8659132/63287573-a882dc80-c288-11e9-83e0-b352e43bdf28.png)

For semantic segmentation, their method is ok, though ENet for example performs better on average and is much faster.

![image](https://user-images.githubusercontent.com/8659132/63287643-d700b780-c288-11e9-9d40-5bcaf695a744.png)

On the other hand, for instance segmentation, their method is much faster than the other while still performing well. Not SOTA on performance, but considering the real-time constraint, it's much better.

# Comments

- Most instance segmentation methods tend to be sluggish and overly complicated. This approach is much more elegant in my opinion.
- If they removed the aggressive down/up sampling, I wonder if they would beat MaskRCNN and PANet.
- I'm not sure what's the point of upsampling the semantic map given that we already have the instance map.

Disclaimer: I am an author

# Intro

Experience replay (ER) and generative replay (GEN) are two effective continual learning strategies. In the former, samples from a stored memory are replayed to the continual learner to reduce forgetting. In the latter, old data is compressed with a generative model and generated data is replayed to the continual learner. Both of these strategies assume a random sampling of the memories. But learning a new task doesn't cause **equal** interference (forgetting) on the previous tasks!  

In this work, we propose a controlled sampling of the replays. Specifically, we retrieve the samples which are most interfered, i.e. whose prediction will be most negatively impacted by the foreseen parameters update. The method is called Maximally Interfered Retrieval (MIR).

## Cartoon for explanation

https://i.imgur.com/5F3jT36.png

Learning about dogs and horses might cause more interference on lions and zebras than on cars and oranges. Thus, replaying lions and zebras would be a more efficient strategy.

# Method

1) incoming data: $(X_t,Y_t)$

2) foreseen parameter update: $\theta^v= \theta-\alpha\nabla\mathcal{L}(f_\theta(X_t),Y_t)$

### applied to ER (ER-MIR)
3) Search for the top-$k$ values $x$ in the stored memories using the criterion $$s_{MI}(x) = \mathcal{L}(f_{\theta^v}(x),y) -\mathcal{L}(f_{\theta}(x),y)$$

### or applied to GEN (GEN-MIR)
3)   
$$
     \underset{Z}{\max} \, \mathcal{L}\big(f_{\theta^v}(g_\gamma(Z)),Y^*\big) -\mathcal{L}\big(f_{\theta}(g_\gamma(Z)),Y^*\big)
$$
$$
         \text{s.t.}   \quad ||z_i-z_j||_2^2 > \epsilon \forall  z_i,z_j \in Z \,\text{with} \, z_i\neq z_j
$$
i.e. search in the latent space of a generative model $g_\gamma$ for samples that are the most forgotten given the foreseen update.

4) Then add theses memories to incoming data $X_t$ and train $f_\theta$

# Results

### qualitative
https://i.imgur.com/ZRNTWXe.png

Whilst learning 8s and 9s (first row), GEN-MIR mainly retrieves 3s and 4s (bottom two rows) which are similar to 8s and 9s respectively.

### quantitative 

GEN-MIR was tested on MNIST SPLIT and Permuted MNIST, outperforming the baselines in both cases.

ER-MIR was tested on MNIST SPLIT, Permuted MNIST and Split CIFAR-10, outperforming the baselines in all cases.


# Other stuff
### (for avid readers)

We propose a hybrid method (AE-MIR) in which the generative model is replaced with an autoencoder to facilitate the compression of harder dataset like e.g. CIFAR-10.

### Summary
Knowing when a model is qualified to make a prediction is critical to safe deployment of ML technology. Model-independent / Unsupervised Out-of-Distribution (OoD) detection is appealing mostly because it doesn't require task-specific labels to train. It is tempting to suggest a simple one-tailed test in which lower likelihoods are OoD (assigned by a Likelihood Model), but the intuition that In-Distribution (ID) inputs should have highest likelihoods _does not hold in higher dimension_. The authors propose to use the Watanabe-Akaike Information Criterion (WAIC) to circumvent this problem and empirically show the robustness of the approach.

### Counterintuitive Properties of Likelihood Models:
https://i.imgur.com/4vo0Ff5.png
So a GLOW model with Gaussian prior maps SVHN closer to the origin than Cifar (but never actually generates SVHN because Gaussian samples are on the shell). This is bad news for OoD detection.

### Proposed Methodology:
Use the WAIC criterion for OoD detection which gives an asymptotically correct estimate of the gap between the training set and test set expectations:
https://i.imgur.com/vasSxuk.png
Basically,  the correction term subtracts the variance in likelihoods across independent samples from the posterior. This acts to robustify the estimate, ensuring that points that are sensitive to the particular choice of posterior are penalized. They use an ensemble of generative models as a proxy for posterior samples i.e. the ensembles acts as approximate posterior samples.
Now, OoD can be detected with a Likelihood Model:
https://i.imgur.com/M3CDKOA.png
### Discussion
Interestingly, GLOW maps Cifar and other datasets INSIDE the gaussian shell (which is an annulus of radius $\sqrt{dim} = \sqrt{3072} \approx 55.4$
https://i.imgur.com/ERdgOaz.png
This is in itself quite disturbing, as it suggests that better flow-based generative models (for sampling) can be obtained by encouraging the training distribution to overlap better with the typical set in latent
space.

Tree-based ML models are becoming increasingly popular, but in the explanation space for these type of models is woefully lacking explanations on a local level. Local level explanations can give a clearer picture on specific use-cases and help pin point exact areas where the ML model maybe lacking in accuracy.

**Idea**: We need a local explanation system for trees, that is not based on simple decision path, but rather weighs each feature in comparison to every other feature to gain better insight on the model's inner workings.

**Solution**: This paper outlines a new methodology using SHAP relative values, to weigh pairs of features to get a better local explanation of a tree-based model. The paper also outlines how we can garner global level explanations from several local explanations, using the relative score for a large sample space. The paper also walks us through existing methodologies for local explanation, and why these are biased toward tree depth as opposed to actual feature importance.

The proposed explanation model titled TreeExplainer exposes methods to compute optimal local explanation, garner global understanding from local explanations, and capture feature interaction within a tree based model.

This method assigns Shapley interaction values to pairs of features essentially ranking the features so as to understand which features have a higher impact on overall outcomes, and analyze feature interaction.

# Summary

The authors present a way to generate captions describing the content of images using attention-based mechanisms. They present two ways of training the network, one via standard backpropagation techniques and another using stochastic processes. They also show how their model can selectively "focus" on the relevant parts of an image to generate appropriate captions, as shown in the classic example of the famous woman throwing a frisbee. Finally, they validate their model on Flicker8k, Flicker30k and MSCOCO.

![image](https://user-images.githubusercontent.com/18450628/61397054-10639300-a897-11e9-8b4a-f4cd804c3229.png)

# Model

At a very high level, the model takes as input an image I and returns a caption generated from a pre-defined vocabulary:

![image](https://user-images.githubusercontent.com/18450628/61398513-20c93d00-a89a-11e9-8e93-72ccf7a61be1.png)

A high-level overview of the model is presented in Figure 1:

![image](https://user-images.githubusercontent.com/18450628/61398365-de076500-a899-11e9-8413-55ec755f0f83.png)

## Visual extractor

A CNN is used to extract features from the image. The authors experimented with VGG-19 pretrained on ImageNet and not finetuned. They use the features from the last convolutional layer as their representations. Starting with images of 224x224, the last CNN feature map has shape 14x14x512, which they flatten along width and height to obtain a vector representation of 196x512. These 512 vectors are used as inputs to the language model.

## Sentence generation

An LSTM network is used to generate a sequence of words from a fixed vocabulary of size L. As input, a weighted sum based on attention values of the vectors of the flattened image features is used. The previous word is also fed as input to the LSTM. The hidden layer from the previous timestep as well as the layers from the CNN a_i are fed through an MLP + softmax layer and used to generate attention values for each flattened image feature vector that sum to one. 

![image](https://user-images.githubusercontent.com/18450628/61462206-41e46900-a940-11e9-991d-e3a9e4b98837.png)

![image](https://user-images.githubusercontent.com/18450628/61462544-c9ca7300-a940-11e9-8c31-dbf85bf8301f.png)




The authors propose two ways to compute phi, i.e. the attention, which they refer to as "soft attention" and "hard attention". These will be covered in a later section.

The output of the LSTM, z, is then fed to a deep network to generate the next word. This is detailed in the following figure.

![image](https://user-images.githubusercontent.com/18450628/61408594-6132b600-a8ae-11e9-894c-392396e299b0.png)


## Attention

The paper proposes two methods of attention, a "soft" attention and a "hard" attention.

### Soft attention

Soft attention is the most intuitive one and is relatively straight forward. In order to compute the vector representing the image as input to the LSTM, **z**, the expectation of the context vector is computed by using a weighted average scheme:

![image](https://user-images.githubusercontent.com/18450628/61408939-34cb6980-a8af-11e9-989e-24308be3ed3c.png)

where alpha are the attention weights and a_i are the vectors of the feature representation.

To ensure that all image features are used somewhat equally, a regularization term is added to the loss function:

![image](https://user-images.githubusercontent.com/18450628/61412057-d2c23280-a8b5-11e9-9d9c-7f35edc650ef.png)

This ensures that the image feature vectors over time sum to 1 as closely as possible and that no part of the image is ignored.

### Hard attention

The authors propose an alternative method to calculate attention. Each attention parameter is treated as an intermediate latent variables that can be represented in one-hot encoding, i.e. on or off. To do so, they use a multinoulli distribution parametrized by alpha, the softmax output of f_att. They show how they can approximate the gradient using monte-carlo methods:

![image](https://user-images.githubusercontent.com/18450628/61413158-d1463980-a8b8-11e9-8aef-b2a9d9bb6bad.png)

Refer to the paper for more mathemagical details. Finally, they use soft attention with probability 0.5 when using hard attention.

## Visualizing features

One of the contributions of this work is showing what the network is "attending" to. To do so, the authors use the final layer of VGG-19, which consists of 14x14x512 features, upsample the resulting filters to the original image size of 224x224 and use a Gaussian blur to recreate the receptive field.



## Results

The authors evaluate their methods on 3 caption datasets, i.e. Flicker8k, Flicker30k and MSCOCO.  For all our experiments, they used a fixed vocabulary size of 10,000. They report both BLEU and METEOR scores on the task.

As can be seen in the following figure, both the soft and hard attention mechanisms beat all state of the art methods at the time of publishing. Hard attention outperformed soft attention most of the time.

![image](https://user-images.githubusercontent.com/18450628/61412281-77dd0b00-a8b6-11e9-882f-73e86638bc9d.png)

# Comments

Cool paper which lead the way in terms of combining text and images and using attention mechanisms. They show an interesting way to visualize what the network is attending to, although it was not clear to me why they should expect the final layer of the CNN to be showing that in the first place since they did not finetune on the datasets they were training on. I would expect that to mean that their method would work best on datasets most "similar" to ImageNet. 

Their hard attention mechanism seems a lot more complicated than the soft attention mechanism and it isn't always clear that it is much better other than it offers a stronger regularization and a type of dropout.

# Summary

This paper presents state-of-the-art methods for both caption generation of images and visual question answering (VQA). The authors build on previous methods by adding what they call a "bottom-up" approach to previous "top-down" attention mechanisms. They show that using their approach they obtain SOTA on both Image captioning (MSCOCO) and the Visual Question and Answering (2017 VQA challenge). They propose a specific network configurations for each. Their biggest contribution is using Faster-R-CNN to retrieve the "important" parts of an image to focus on in both models.

## Top Down

Up until this paper, the traditional approach was to use a "top-down" approach, in which the last feature map layer of a CNN is used to obtain a latent representation of the given input image. These features, along with the context of the caption being generated, were used to generate attention weights that were used to predict the next sequence in the context of caption generation. The network would learn to focus its attention on regions of the feature map that matters most. This is the approach used in previous SOTA methods like [Show, Attend and Tell: Neural Image Caption Generation with Visual Attention](https://arxiv.org/abs/1502.03044).

## Bottom-up

The authors argue that the feature map of a CNN is too generic and can be thought of operating on a uniform, grid-like feature map. In other words, there is no particular reason to think that the feature map of generated by a CNN would give optimal regions to attend to. Also, carefully choosing the dimensions of the feature map can be very arbitrary.

In order to fix this, the authors propose combining object detection methods in a *bottom-up* approach. To do so, the authors propose using Faster-R-CNN to identify regions of interest in an image. Given an input image, Faster-R-CNN will identify bounding boxes of the image that likely correspond to objects of a given category and simultaneously compute a feature vector of that bounding box. Figure 1 shows the difference between the Bottom-up and Top-Down approach.

![image](https://user-images.githubusercontent.com/18450628/61817263-2683cd00-ae1c-11e9-971a-d3b531dbbd98.png)

## Combining the two

In this paper, the authors suggest using the bottom-up approach to compute the salient regions of the image the network should focus on using Faster-R-CNN. FRCNN is carefully pretrained on both imagenet and the Visual Genome dataset. It is then frozen and only used to generate bounding boxes of regions with high confidence of being of interest. The top-down approach is then used on the features obtained from the bottom-up approach. In order to "enhance" the FRCNN performance, they initialize their FRCNN with a ResNet-101 pre-trained on imagenet. They train their FRCNN on the Visual Genome dataset, adding attributes to the loss function that are available from the Visual Genome dataset, attributes such as color (black, white, gold etc.),  state (open, close, dark, bright, etc.). A sample of FRCNN outputs are shown in figure 2. It is important to stress that only the feature representations and not the actual outputs (i.e. not the labels) are used in their model.

![image](https://user-images.githubusercontent.com/18450628/61817487-aca01380-ae1c-11e9-90fa-134033b95bb0.png)

## Caption Generation

Figure 3 provides a high-level overview of the model being used for caption generation for images. The image is first passed through FRCNN which produces a set of image features *V*. In their specific implementation, *V* consists of *k* vectors of size 1x2048. Their model consists of two LSTM blocks, one for attention and the other for language generation.

![image](https://user-images.githubusercontent.com/18450628/61818488-effb8180-ae1e-11e9-8ae4-14355115429a.png)

The first block of their model is a Top-Down Attention LSTM layer. It takes as input the mean-pooled features *V* , i.e. 1/k*sum(v_i), concatenated with the previous timestep's hidden representation of the language LSTM as well as the word embedding of the previously generated word. The word embedding is learned and not pretrained. 

The output of the first LSTM is used to compute the attention for each vector using an MLP and softmax:

![image](https://user-images.githubusercontent.com/18450628/61819982-21298100-ae22-11e9-80a9-99640896413d.png)

The attention weighted image feature is then used as an input to the language LSTM model, concatenated with the output from the top-down Attention LSTM and a softmax is used to predict the next word in the sequence. The loss function seeks to minimize the cross-entropy of the generated sentence. 

## VQA Model

The VQA task differs to the image generation in that a text-based question accompanies an input image and the network must produce an answer. The VQA model proposed is different to that of the caption generation model previously described, however they both use the same bottom-up approach to generate the feature vectors of the image based on the FRCNN architecture. A high-level overview of the architecture for the VQA model is presented in Figure 4.

![image](https://user-images.githubusercontent.com/18450628/61821988-8da67f00-ae26-11e9-8456-3c9e5ec60787.png)

Each word from the question is converted to a learned word embedding which is used as input to a GRU. The number of words for each question is limited to 14 for computational efficiency. The output from the GRU is concatenated with each of the *k* image features, and attention weights are computed for each *k*th feature using an MLP and softmax, similar to what is done in the attention for caption generation. The weighted sum of the feature vectors is then passed through an linear layer such that its shape is compatible with the gru output, and the Hadamard product (element-wise product) is computed over the GRU output and attention-weighted image feature representation. Finally, a tanh non-linear activation is used. This results in a "gated tanh", which have been shown empirically to outperform both ReLU and tanh. Finally, a softmax probability distribution is generated at the output which selects a candidate answer among all possible candidate answers.

## Results and experiments

### Resnet Baseline

To demonstrate that their contribution of bottom-up mechanism actually improves on results, the authors use a ResNet trained on imagenet as a baseline for generating the image feature vectors (they resize the final CNN layers using bilinear interpolation when needed). They consistently obtain better results when using the bottom-up approach over the ResNet approach in both caption generation and VQA.

## MSCOCO

The authors demonstrate that they outperform all results on all metrics on the MSCOCO test server.

![image](https://user-images.githubusercontent.com/18450628/61824157-4f5f8e80-ae2b-11e9-8d90-657db453e26e.png)

They also show how using the bottom-up approach over ResNet consistently scores them higher on detecting instances of objects, attributes, relations, etc:

![image](https://user-images.githubusercontent.com/18450628/61824238-7fa72d00-ae2b-11e9-81b3-b5a7f80153f3.png)

The authors, like their predecessors, insist on demonstrating their network's frisbee ability:

![image](https://user-images.githubusercontent.com/18450628/61824344-bed57e00-ae2b-11e9-87cd-597568587e1d.png)

## VQA Results

They also demonstrate that the addition of bottom-up attention improves results over a ResNet baseline. 

![image](https://user-images.githubusercontent.com/18450628/61824500-28ee2300-ae2c-11e9-9016-2120a91917e4.png)

They also show that their model outperformed all other submissions on the VQA submission. They mention using an ensemble of 30 models for their submission. 

![image](https://user-images.githubusercontent.com/18450628/61824634-83877f00-ae2c-11e9-8d84-9589e0ea2be2.png)


A sample image of what is attended in an image given a proper answer is shown in figure 6.

![image](https://user-images.githubusercontent.com/18450628/61824608-736f9f80-ae2c-11e9-9d4e-8cb6bd0a1a92.png)


# Comments

The authors introduce a new way to select portions of the image on which to focus attention. The idea is very original and came at a time when object detection was making significant progress (i.e. FRCNN).

A few comments:

* This method might not generalize well to other types of data. It requires pre-training on larger datasets (visual genome, imagenet, etc.) which consist of categories that overlap with both the MSCOCO and VQA datasets (i.e. cars, people, etc.). It would be interesting to see an end-to-end model that does not rely on pre-training on other similar datasets.

* No insight is given to the computational complexity nor to the inference time or training time. I imagine that FRCNN is resource intensive, and having to do a forward pass of FRCNN for every pass of the network must be a computational bottleneck. Not to mention that they ensembled 30 of them!

The idea is to combine collaborative filtering with content-based recommenders to mitigate the user and item coldstart problems.

The author distinguishes between positive and negative interactions.

The representation of a user and of items is the sum of all their latent representations. This sounds similar to "**Asymmetric factor models**" as described in [the BellKor Netflix price solution](https://www.netflixprize.com/assets/ProgressPrize2007_KorBell.pdf). **The key idea is to encode the latent user (or item) vector as a sum of latent attribute vectors.**

Adagrad / asynchronous stochastic gradient descent was used for optimization.


## See also

* [Code on GitHub](https://lyst.github.io/lightfm/docs/index.html#)
* [Paper on ArXiv](https://arxiv.org/pdf/1507.08439.pdf)

This paper is about a recommendation system approach using collaborative filtering (CF) on implicit feedback datasets.

The core of it is the minimization problem

$$\min_{x_*, y_*} \sum_{u,i} c_{ui} (p_{ui} - x_u^T y_i)^2 + \underbrace{\lambda \left ( \sum_u || x_u ||^2 + \sum_i || y_i ||^2\right )}_{\text{Regularization}}$$

with

* $\lambda \in [0, \infty[$ is a hyper parameter which defines how strong the model is regularized
* $u$ denoting a user, $u_*$ are all user factors $x_u$ combined
* $i$ denoting an item, $y_*$ are all item factors $y_i$ combined
* $x_u \in \mathbb{R}^n$ is the latent user factor (embedding); $n$ is another hyper parameter. $n=50$ seems to be a reasonable choice.
* $y_i \in \mathbb{R}^n$ is the latent item factor (embedding)
* $r_{ui}$ defines the "intensity"; higher values mean user $u$ interacted more with item $i$
* $p_{ui} = \begin{cases}1 & \text{if } r_{ui} >0\\0 &\text{otherwise}\end{cases}$
* $c_{ui} := 1 + \alpha r_{ui}$ where $\alpha \in [0, \infty[$ is a hyper parameter; $\alpha =40$ seems to be reasonable

In contrast, the standard matrix factoriation optimization function looks like this ([example](https://www.cs.cmu.edu/~mgormley/courses/10601-s17/slides/lecture25-mf.pdf)):

$$\min_{x_*, y_*} \sum_{(u, i, r_{ui}) \in \mathcal{R}} {(r_{ui} - x_u^T y_i)}^2  + \underbrace{\lambda \left ( \sum_u || x_u ||^2 + \sum_i || y_i ||^2\right )}_{\text{Regularization}}$$

where

* $\mathcal{R}$ is the set of all ratings $(u, i, r_{ui})$ - user $u$ has rated item $i$ with value $r_{ui} \in \mathbb{R}$

They use alternating least squares (ALS) to train this model.

The prediction then is the dot product between the user factor and all item factors ([source](https://github.com/benfred/implicit/blob/master/implicit/recommender_base.pyx#L157-L176))

The paper designs some basic tests to compare saliency methods. It founds that some of the most popular methods are independent of model parameters and the data, meaning they are effectively useless.

## Methods compared

The paper compare the following methods: gradient explanation, gradient x input, integrated gradients, guided backprop, guided GradCam and SmoothGrad. They provide a refresher on those methods in the appendix.

All those methods can be put in the same framework. They require a classification model and an input (typically an image). The output of the method is an *explanation map* of the shape of the input where a higher value for a feature implies greater relevance in the decision of the model. 

## Metrics of comparison

The authors argue that visual inspection of the saliency maps can be misleading. They propose to compute the Spearman rank correlation, the structural similarity index (SSMI) and the Pearson correlation of the histogram of gradients. The authors point out that those metrics capture various notions of similarity, but it is an active area of research and those metrics are imperfect.

## First test: model parameters randomization

A saliency method must be dependent of model parameters, otherwise it cannot help us understand a model. In this test, the authors randomize the model parameters, layer per layer, starting from the top.

Surprisingly, methods such as guided backprop and guided gradcam are completely insensitive to model parameters, as illustrated on this Inception v3 trained on ImageNet:

![image](https://user-images.githubusercontent.com/8659132/61403152-b10b8000-a8a2-11e9-9f6a-cf1ed6a876cc.png)

Integrated gradients looks also dubious as the bird is still visible with a mostly fully randomized model, but the quantitative metrics reveal the difference is actually big between the two models.

## Second test: data randomization

It is well-known that randomly shuffling the labels of a dataset does not prevent a neural network from getting a high accuracy on the training set, though it does prevent generalization. The model is able to learn by either memorizing the data or finding spurious patterns. As a result, saliency maps obtained from such a network should have no clearly interpretable signal.

Here is the result for a ConvNet trained on MNIST and a shuffled MNIST:

![image](https://user-images.githubusercontent.com/8659132/61406757-7efe1c00-a8aa-11e9-9826-a859a373cb4f.png)

The results are very damning for most methods. Only gradients and GradCam are very different between both models, as confirmed by the low correlation.

## Discussion

- Even though some methods do no depend on model parameters and data, they might still depend on the architecture of the models, which could be of some use in some contexts.
- Methods that multiply the input with the gradient are dominated by the input.
- Complex saliency methods are just fancy edge detectors.
- Only gradient, smooth gradient and GradCam survives the sanity checks.

# Comments

- Why is their GradCam maps so ugly? They don't look like usual GradCam maps at all.
- Their tests are simple enough that it's hard to defend a method that doesn't pass them.
- The methods that are left are not very good either. They give fuzzy maps that are difficult to interpret.
- In the case of integrated gradients (IG), I'm not convinced this is sufficient to discard the method. IG requires a "baseline input" that represents the absence of features. In the case of images, people usually just set the image to 0, which is not at all the absence of a feature. The authors also use the "set the image to 0" strategy, and I'd say their tests are damning for this strategy, not for IG in general. I'd expect an estimation of the baseline such as done in [this paper](https://arxiv.org/abs/1702.04595) would be a fairer evaluation of IG.

Code: [GitHub](https://github.com/adebayoj/sanity_checks_saliency) (not available as of 17/07/19)

Xu and Mannor provide a theoretical paper on robustness and generalization where their notion of robustness is based on the idea that the difference in loss should be small for samples that are close. This implies that, e.g., for a test sample close to a training sample, the loss on both samples should be similar. The authors formalize this notion as follows:

Definition: Let $A$ be a learning algorithm and $S \subset Z$ be a training set such that $A(S)$ denotes the model learned on $S$ by $A$; the algorithm $A$ is $(K, \epsilon(S))$-robust if $Z$ can be partitioned into $K$ disjoint sets, denoted $C_i$ such that $\forall s \in S$ it holds:

$s,z \in C_i \rightarrow |l(A(S), s) – l(A(S), z)| \leq \epsilon(S)$.

In words, this means that we can partition the space $Z$ (which is $X \times Y$ for a supervised problem) into a finite set of subsets and whenever a sample falls into the same partition as a training sample, the learned model should have nearly the same loss on both samples. Note that this notion does not entirely match the notion of adversarial robustness as commonly referred to nowadays. The main difference is that the partition can be chosen, while for adversarial robustness, the “partition” (usually in form of epsilon-balls around training and testing samples) is fixed.

Based on the above notion of robustness, the authors provide PAC bounds for robust algorithms, i.e. generalization performance of $A$ is linked to its generalization. Furthermore, in several examples, common machine learning techniques such as SVMs and neural networks are shown to be robust under specific conditions. For neural networks, for example, an upper bound on the $L_1$ norm of weights and the requirement of Lipschitz continuity is enough. This actually related to work on adversarial robustness, where Lipschitz continuity and weight regularization is also studied.

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