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.

This is an interestingly pragmatic paper that makes a super simple observation. Often, we may want a usable network with fewer parameters, to make our network more easily usable on small devices. It's been observed (by these same authors, in fact), that pruned networks can achieve comparable weights to their fully trained counterparts if you rewind and retrain from early in the training process, to compensate for the loss of the (not ultimately important) pruned weights. This observation has been dubbed the "Lottery Ticket Hypothesis", after the idea that there's some small effective subnetwork you can find if you sample enough networks. 

Given these two facts - the usefulness of pruning, and the success of weight rewinding - the authors explore the effectiveness of various ways to train after pruning. Current standard practice is to prune low-magnitude weights, and then continue training remaining weights from values they had at pruning time, keeping the final learning rate of the network constant. The authors find that: 
1. Weight rewinding, where you rewind weights to *near* their starting value, and then retrain using the learning rates of early in training, outperforms fine tuning from the place weights were when you pruned 

but, also 
2. Learning rate rewinding, where you keep weights as they are, but rewind learning rates to what they were early in training, are actually the most effective for a given amount of training time/search cost 

To me, this feels a little bit like burying the lede: the takeaway seems to be that when you prune, it's beneficial to make your network more "elastic" (in the metaphor-to-neuroscience sense) so it can more effectively learn to compensate for the removed neurons. So, what was really valuable in weight rewinding was the ability to "heat up" learning on a smaller set of weights, so they could adapt more quickly. And the fact that learning rate rewinding works better than weight rewinding suggests that there is value in the learned weights after all, that value is just outstripped by the benefit of rolling back to old learning rates. 

All in all, not a super radical conclusion, but a useful and practical one to have so clearly laid out in a paper.

Offline reinforcement learning is potentially high-value thing for the machine learning community learn to do well, because there are many applications where it'd be useful to generate a learnt policy for responding to a dynamic environment, but where it'd be too unsafe or expensive to learn in an on-policy or online way, where we continually evaluate our actions in the environment to test their value. In such settings, we'd like to be able to take a batch of existing data - collected from a human demonstrator, or from some other algorithm - and be able to learn a policy from those pre-collected transitions, without being able to query the environment further by taking arbitrary actions. 

There are two broad strategies for learning a policy from precollected transitions. One is to simply learn to mimic the action policy used by the demonstrator, predicting the action the demonstrator would take in a given state, without making use of reward data at all. This is Behavioral Cloning, and has the advantage of being somewhat more conservative (in terms of not experimenting with possibly-unsafe-or-low-reward actions the demonstrator never took), but this is also a disadvantage, because it's not possible to get higher reward than the demonstrator themselves got if you're simply copying their behavior. Another approach is to learn a Q function - estimating the value of a given action in a given state - using the reward data from the precollected transitions. This can also have some downsides, mostly in the direction of overconfidence. Q value Temporal Difference learning works by using the current reward added to the max Q value over possible next actions as the target for the current-state Q estimate. This tends to lead to overestimates, because regression to the mean effects mean that the highest value Q estimates are disproportionately likely to be noisy (possibly because they correspond to an action with little data in the demonstrator dataset). In on-policy Q learning, this is less problematic, because the agent can take the action associated with their noisily inaccurate estimate, and as a result get more data for that action, and get an estimate that is less noisy in future. But when we're in a fully offline setting, all our learning is completed before we actually start taking actions with our policy, so taking high-uncertainty actions isn't a valuable source of new information, but just risky. 

The approach suggested by this DeepMind paper - Critic Regularized Regression, or CRR - is essentially a synthesis of these two possible approaches. The method learns a Q function as normal, using temporal difference methods. The distinction in this method comes from how to get a policy, given a learned Q function. Rather than simply taking the action your Q estimate says is highest-value at a particular point, CRR optimizes a policy according to the formula shown below. The f() function is a stand-in for various potential functions, all of which are monotonic with respect to the Q function, meaning they increase when the Q function does. 

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

This basically amounts to a form of a behavioral cloning loss (with the part that maximizes the probability under your policy of the actions sampled from the demonstrator dataset), but weighted or, as the paper terms it, filtered, by the learned Q function. The higher the estimated q value for a transition, the more weight is placed on that transition from the demo dataset having high probability under your policy. Rather than trying to mimic all of the actions of the demonstrator, the policy preferentially tries to mimic the demonstrator actions that it estimates were particularly high-quality. Different f() functions lead to different kinds of filtration. The `binary`version is an indicator function for the Advantage of an action (the Q value for that action at that state minus some reference value for the state, describing how much better the action is than other alternatives at that state) being greater than zero. Another, `exp`, uses exponential weightings which do a more "soft" upweighting or downweighting of transitions based on advantage, rather than the sharp binary of whether an actions advantage is above 1. 

The authors demonstrate that, on multiple environments from three different environment suites, CRR outperforms other off-policy baselines - either more pure behavioral cloning, or more pure RL - and in many cases does so quite dramatically. They find that the sharper binary weighting scheme does better on simpler tasks, since the trade-off of fewer but higher-quality samples to learn from works there. However, on more complex tasks, the policy benefits from the exp weighting, which still uses and learns from more samples (albeit at lower weights), which introduces some potential mimicking of lower-quality transitions, but at the trade of a larger effective dataset size to learn from.

This is a mildly silly paper to summarize, since there isn't really a new mechanism to understand, but rather a number of straightforward (and interesting!) empirical results that are also quite well-explained in the paper itself. That said, for the sake of a tiny bit more brevity than the paper itself provides, I'll try to pull out some of the conclusions I found the most interesting here. 

The general goal of this paper is to better understand the contours of when self-supervised representation learning is valuable for vision (and specifically when it can compete with supervised learning), and when it doesn't. In general, the results are all using ResNet backbones, with SimCLR SSL, on image classification datasets. Some bullet-point takeaways: 

- The SSL models being tested here seem to roughly saturate at unsupervised dataset sizes of around 500K; the comparative jump from dataset sizes of 500K to 1M is fairly small.
- Once you have a supervised dataset of around 50K or more, the benefit of SSL pretraining starts to diminish, and it converges to being more similar to just supervised learning on that numbrer of labeled images. On the flip side, it's only possible to get close to "good" fully supervised performance by using 100K images or more on top of a SSL baseline.
- Even within image classification datasets, it's much better to do SSL representation on the same dataset as the one you'll use for downstream training; trying to transfer representations to different datasets leads to meaningfully worse results. Interestingly, this is even true when you add out-of-domain (i.e. other-dataset) data to an existing in-domain dataset: a dataset of 250K in-dataset images does better than a 500K dataset of images from mixed datasets, and does notably better than a 1M dataset of mixed images. In this case, adding more out-of-domain images seems to have just degraded performance
- SSL seems to perform more closely to SL on a course label set; when the label set gets more granular, the task gets harder overall, but, more specifically, the gap between SSL and SL grows
- When the authors tried different forms of dataset corruption, SSL was much more robust to adding salt-and-pepper noise than it was to removing high-frequency information in the form of reducing the images to a lower resolution.

Kumar et al. propose an algorithm to learn in batch reinforcement learning (RL), a setting where an agent learns purely form a fixed batch of data, $B$, without any interactions with the environments. The data in the batch is collected according to a batch policy $\pi_b$. Whereas most previous methods (like BCQ) constrain the learned policy to stay close to the behavior policy, Kumar et al. propose bootstrapping error accumulation reduction (BEAR), which constrains the newly learned policy to place some probability mass on every non negligible action. 
The difference is illustrated in the picture from the BEAR blog post:
https://i.imgur.com/zUw7XNt.png
The behavior policy is in both images the dotted red line, the left image shows the policy matching where the algorithm is constrained to the purple choices, while the right image shows the support matching. 

**Theoretical Contribution:**
The paper analysis formally how the use of out-of-distribution actions to compute the target in the Bellman equation influences the back-propagated error.

Firstly a distribution constrained backup operator is defined as
$T^{\Pi}Q(s,a) = \mathbb{E}[R(s,a) + \gamma \max_{\pi \in \Pi} \mathbb{E}_{P(s' \vert s,a)} V(s')]$ and $V(s) = \max_{\pi \in \Pi} \mathbb{E}_{\pi}[Q(s,a)]$
which considers only policies $\pi \in \Pi$. 

It is possible that the optimal policy $\pi^*$ is not contained in the policy set $\Pi$, thus there is a suboptimallity constant $\alpha (\Pi) = \max_{s,a} \vert \mathcal{T}^{\Pi}Q^{*}(s,a) - \mathcal{T}Q^{*}(s,a) ]\vert $ which captures how far $\pi^{*}$ is from $\Pi$. 

Letting $P^{\pi_i}$ be the transition-matrix when following policy $\pi_i$, $\rho_0$ the state marginal distribution of the training data in the batch and $\pi_1, \dots, \pi_k \in \Pi $. The error analysis relies upon a concentrability assumption $\rho_0 P^{\pi_1} \dots P^{\pi_k} \leq c(k)\mu(s)$, with $\mu(s)$ the state marginal. Note that $c(k)$ might be infinite if the support of $\Pi$ is not contained in the state marginal of the batch. Using the coefficients $c(k)$ a concentrability coefficient is defined as:
$C(\Pi) = (1-\gamma)^2\sum_{k=1}^{\infty}k \gamma^{k-1}c(k).$
The concentrability takes values between 1 und $\infty$, where 1 corresponds to the case that the batch data were collected by $\pi$ and $\Pi = \{\pi\}$ and $\infty$ to cases where $\Pi$ has support outside of $\pi$. 

Combining this Kumar et a. get a bound of the Bellman error for distribution constrained value iteration with the constrained Bellman operator $T^{\Pi}$:

$\lim_{k \rightarrow \infty} \mathbb{E}_{\rho_0}[\vert V^{\pi_k}(s)- V^{*}(s)]  \leq  \frac{\gamma}{(1-\gamma^2)} [C(\Pi) \mathbb{E}_{\mu}[\max_{\pi \in \Pi}\mathbb{E}_{\pi}[\delta(s,a)] + \frac{1-\gamma}{\gamma}\alpha(\Pi)  ] ]$, where $\delta(s,a)$ is the Bellman error. 

This presents the inherent batch RL trade-off between keeping policies close to the behavior policy of the batch (captured by $C(\Pi)$ and keeping $\Pi$ sufficiently large (captured by $\alpha(\Pi)$). 

It is finally proposed to use support sets to construct $\Pi$, that is $\Pi_{\epsilon} = \{\pi \vert \pi(a \vert s)=0 \text{ whenever } \beta(a \vert s) < \epsilon \}$. This amounts to the set of all policies that place probability on all non-negligible actions of the behavior policy. For this particular choice of $\Pi = \Pi_{\epsilon}$ the concentrability coefficient can be bounded. 

**Algorithm**:
The algorithm has an actor critic style, where the Q-value to update the policy is taken to be the minimum over the ensemble. The support constraint to place at least some probability mass on every non negligible action from the batch is enforced via sampled MMD. The proposed algorithm is a member of the policy regularized algorithms as the policy is updated to optimize:
$\pi_{\Phi} = \max_{\pi} \mathbb{E}_{s \sim B} \mathbb{E}_{a \sim \pi(\cdot \vert s)} [min_{j = 1 \dots, k} Q_j(s,a)] s.t. \mathbb{E}_{s \sim B}[MMD(D(s), \pi(\cdot \vert s))] \leq \epsilon$

The Bellman target to update the Q-functions is computed as the convex combination of minimum and maximum of the ensemble. 

**Experiments**
The experiments use the Mujoco environments Halfcheetah, Walker, Hopper and Ant. Three scenarios of batch collection, always consisting of 1Mio. samples,  are considered:
- completely random behavior policy
- partially trained behavior policy
- optimal policy as behavior policy

The experiments confirm that BEAR outperforms other off-policy methods like BCQ or KL-control. The ablations show further that the choice of MMD is crucial as it is sometimes on par and sometimes substantially better than choosing KL-divergence.