This paper posits that one of the central problems stopping multi-task RL - that is, single models trained to perform multiple tasks well - from reaching better performance, is the inability to balance model resources and capacity between the different tasks the model is being asked to learn. Empirically, prior to this paper, multi-task RL could reach ~50% of human accuracy on Atari and Deepmind Lab tasks. The fact that this is lower than human accuracy is actually somewhat less salient than the fact that it’s quite a lot lower than single-task RL - how a single model trained to perform only that task could do. 

When learning a RL model across multiple tasks, the reward structures of the different tasks can vary dramatically. Some can have high-magnitude, sparse rewards, some can have low magnitude rewards throughout. If a model learns it can gain what it thinks is legitimately more reward by getting better at a game with an average reward of 2500 than it does with an average reward of 15, it will put more capacity into solving the former task. Even if you apply normalization strategies like reward clipping (which treats all rewards as a binary signal, regardless of magnitude, and just seeks to increase the frequency of rewards), that doesn’t deal with some environments having more frequent rewards than others, and thus more total reward when summed over timestep. 


The authors here try to solve this problem by performing a specific kind of normalization, called Pop Art normalization, on the problem. PopArt normalization (don’t worry about the name) works by adaptively normalizing both the target and the estimate of the target output by the model, at every step. In the Actor-Critic case that this model is working on, the target and estimate that are being normalized are, respectively, 1) the aggregated rewards of the trajectories from state S onward, and 2) the value estimate at state S. If your value function is perfect, these two things should be equivalent, and so you optimize your value function to be closer to the true rewards under your policy. And, then, you update your policy to increase probability of actions with higher advantage (expected reward with that action, relative to the baseline Value(S) of that state).  The “adaptive” part of that refers to correcting for the fact when you’re estimating, say, a Value function to predict the total future reward of following a policy at a state, that V(S) will be strongly non-stationary, since by improving your policy you are directly optimizing to increase that value. This is done by calculating “scale” and “shift” parameters off of a recent data. 

The other part of the PopArt algorithm works by actually updating the estimate our model is producing, to stay normalized alongside the continually-being-re-normalized target. 

https://i.imgur.com/FedXTfB.png
It does this by taking the new and old versions of scale (sigma) and shift (mu) parameters (which will be used to normalize the target) and updates the weights and biases of the last layer, such that the movement of the estimator moves along with the movement in the target. 

Using this toolkit, this paper proposes learning one *policy* that’s shared over all task, but keeping shared value estimation functions for each task. Then, it normalizes each task’s values independently, meaning that each task ends up contributing equal weight to the gradient updates of the model (both for the Value and Policy updates). In doing this, the authors find dramatically improved performance at both Atari and Deepmind, relative to prior IMPALA work
https://i.imgur.com/nnDcjNm.png
https://i.imgur.com/Z6JClo3.png

Federated learning is the problem of training a model that incorporates updates from the data of many individuals, without having direct access to that data, or having to store it. This is potentially desirable both for reasons of privacy (not wanting to have access to private data in a centralized way), and for potential benefits to transport cost when data needed to train models exists on a user's device, and would require a lot of bandwidth to transfer to a centralized server.  

Historically, the default way to do Federated Learning was with an algorithm called FedSGD, which worked by: 

- Sending a copy of the current model to each device/client
- Calculating a gradient update to be applied on top of that current model given a batch of data sampled from the client's device
- Sending that gradient back to the central server
- Averaging those gradients and applying them all at once to a central model

The authors note that this approach is equivalent to one where a single device performs a step of gradient descent locally, sends the resulting *model* back to the the central server, and performs model averaging by averaging the parameter vectors there. Given that, and given their observation that, in federated learning, communication of gradients and models is generally much more costly than the computation itself (since the computation happens across so many machines), they ask whether the communication required to get to a certain accuracy could be better optimized by performing multiple steps of gradient calculation and update on a given device, before sending the resulting model back to a central server to be average with other clients models. 

Specifically, their algorithm, FedAvg, works by: 

- Dividing the data on a given device into batches of size B
- Calculating an update on each batch and applying them sequentially to the starting model sent over the wire from the server
- Repeating this for E epochs

Conceptually, this should work perfectly well in the world where data from each batch is IID - independently drawn from the same distribution. But that is especially unlikely to be true in the case of federated learning, when a given user and device might have very specialized parts of the data space, and prior work has shown that there exist pathological cases where averaged models can perform worse than either model independently, even *when* the IID condition is met. 

The authors experiment empirically ask the question whether these sorts of pathological cases arise when simulating a federated learning procedure over MNIST and a language model trained on Shakespeare, trying over a range of hyperparameters (specifically B and E), and testing the case where data is heavily non-IID (in their case: where different "devices" had non-overlapping sets of digits). 

https://i.imgur.com/xq9vi8S.png

They show that, in both the IID and non-IID settings, they are able to reach their target accuracy, and are able to do so with many fewer rounds of communciation than are required by FedSGD (where an update is sent over the wire, and a model sent back, for each round of calculation done on the device.) The authors argue that this shows the practical usefulness of a Federated Learning approach that does more computation on individual devices before updating, even in the face of theoretical pathological cases.

The paper discusses and empirically investigates by empirical testing the effect of "catastrophic forgetting" (**CF**), i.e. the inability of a model to perform a task it was previously trained to perform if retrained to perform a second task. 

An illuminating example is what happens in ML systems with convex objectives: regardless of the initialization (i.e. of what was learnt by doing the first task), the training of the second task will always end in the global minimum, thus totally "forgetting" the first one. 

Neuroscientific evidence (and common sense) suggest that the outcome of the experiment is deeply influenced by the similarity of the tasks involved. Namely, if (i) the two tasks are *functionally identical but input is presented in a different format* or if (ii)  *tasks are similar* and the third case for (iii) *dissimilar tasks*. 

Relevant examples may be provided respectively by (i) performing the same image classification task starting from two different image representations as RGB or HSL, (ii) performing image classification tasks with semantically similar as classifying two similar animals and (iii) performing a text classification followed by image classification. 

The problem is investigated by an empirical study covering two methods of training ("SGD" and "dropout") combined with 4 activations functions (logistic sigmoid, RELU, LWTA, Maxout). A random search is carried out on these parameters. 

From a practitioner's point of view, it is interesting to note that dropout has been set to 0.5 in hidden units and 0.2 in the visible one since this is a reasonably well-known parameter. 

## Why the paper is important
It is apparently the first to provide a systematic empirical analysis of CF. Establishes a framework and baselines to face the problem.

## Key conclusions, takeaways and modelling remarks
* dropout helps in preventing CF
* dropout seems to increase the optimal model size with respect to the model without dropout 
* choice of activation function has a less consistent effect than dropout\no dropout choice
* dissimilar task experiment provides a  notable exception of then dissimilar task experiment
* the previous hypothesis that LWTA activation is particularly resistant to CF is rejected  (even if it performs best in the new task in the dissimilar task pair the behaviour is inconsistent)
* choice of activation function should always be cross-validated
* If computational resources are insufficient for cross-validation the combination dropout + maxout activation function is recommended.

Sinha et al. introduce a variant of adversarial training based on distributional robust optimization. I strongly recommend reading the paper for understanding the introduced theoretical framework. The authors also provide guarantees on the obtained adversarial loss – and show experimentally that this guarantee is a realistic indicator. The adversarial training variant itself follows the general strategy of training on adversarially perturbed training samples in a min-max framework. In each iteration, an attacker crafts an adversarial examples which the network is trained on. In a nutshell, their approach differs from previous ones (apart from the theoretical framework) in the used attacker. Specifically, their attacker optimizes

$\arg\max_z l(\theta, z) - \gamma \|z – z^t\|_p^2$

where $z^t$ is a training sample chosen randomly during training.

On a side note, I also recommend reading the reviews of this paper: https://openreview.net/forum?id=Hk6kPgZA-

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

- Presents a simple visualization method based on “filter normalization.”
- Observed that __the deeper networks become, neural loss landscapes become more chaotic__; causes a dramatic drop in generalization error, and ultimately to a lack of trainability.
- Observed that __skip connections promote flat minimizers and prevent the transition to chaotic behavior__; helps explain why skip connections are necessary for training extremely deep networks.
- Quantitatively measures non-convexity.
- Studies the visualization of SGD optimization trajectories.