Rakelly et al. propose a method to do off-policy meta reinforcement learning (rl). The method achieves a 20-100x improvement on sample efficiency compared to on-policy meta rl like MAML+TRPO. 

The key difficulty for offline meta rl arises from the meta-learning assumption, that meta-training and meta-test time match. However during test time the policy has to explore and sees as such on-policy data which is in contrast to the off-policy data that should be used at meta-training. The key contribution of PEARL is an algorithm that allows for online task inference in a latent variable at train and test time, which is used to train a Soft Actor Critic, a very sample efficient off-policy algorithm, with additional dependence of the latent variable.

The implementation of Rakelly et al. proposes to capture knowledge about the current task in a latent stochastic variable Z. A inference network $q_{\Phi}(z \vert c)$ is used to predict the posterior over latents given context c of the current task in from of transition tuples $(s,a,r,s')$ and trained with an information bottleneck. Note that the task inference is done on samples according to a sampling strategy sampling more recent transitions. The latent z is used as an additional input to policy $\pi(a \vert s, z)$ and Q-function $Q(a,s,z)$ of a soft actor critic algorithm which is trained with offline data of the full replay buffer. 

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


So the challenge of differing conditions at test and train times is resolved by sampling the content for the latent context variable at train time only from very recent transitions (which is almost on-policy) and at test time by construction on-policy. Sampling $z \sim q(z \vert c)$ at test time allows for posterior sampling of the latent variable, yielding efficient exploration. 

The experiments are performed across 6 Mujoco tasks with ProMP, MAML+TRPO and $RL^2$ with PPO as baselines. They show:
- PEARL is 20-100x more sample-efficient 
- the posterior sampling of the latent context variable enables deep exploration that is crucial for sparse reward settings
- the inference network could be also a RNN, however it is crucial to train it with uncorrelated transitions instead of trajectories that have high correlated transitions
- using a deterministic latent variable, i.e. reducing $q_{\Phi}(z \vert c)$ to a  point estimate, leaves the algorithm unable to solve sparse reward navigation tasks which is attributed to the lack of temporally extended exploration. 

The paper introduces an algorithm that allows to combine meta learning with an off-policy algorithm that dramatically increases the sample-efficiency compared to on-policy meta learning approaches. This increases the chance of seeing meta rl in any sort of real world applications. 

Contrastive learning works by performing augmentations on a batch of images, and training a network to match the representations of the two augmented parts of a pair together, and push the representations of images not in a pair farther apart. Historically, these algorithms have benefitted from using stronger augmentations, which has the effect of making the two positive elements in a pair more visually distinct from one another. This paper tries to build on that success, and, beyond just using a strong augmentation, tries to learn a way to perturb images that adversarially increases contrastive loss. As with adversarial training in normal supervised setting, the thinking here is that examples which push loss up the highest are the hardest and thus most informative for the network to learn from 

While the concept of this paper made some sense, I found the notation and the explanation of mechanics a bit confusing, particularly when it came to choice to frame a contrastive loss as a cross-entropy loss, with the "weights" of the dot product in the the cross-entropy loss being, in fact, the projection by the learned encoder of various of the examples in the batch. 

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

This notion of the learned representations being "weights" is just odd and counter-intuitive, and the process of trying to wrap my mind around it isn't one I totally succeeded at. I think the point of using this frame is because it provides an easy analogue to the Fast Gradient Sign Method of normal supervised learning adversarial examples, even though it has the weird effect that, as the authors say "your weights vary by batch...rather than being consistent across training," 

Notational weirdness aside, my understanding is that the method of this paper: 

- Runs a forward pass of normal contrastive loss (framed as cross-entropy loss) which takes augmentations p and q and runs both forward through an encoder.
- Calculates a delta to apply to each input image in the q that will increase the loss most, taken over all the images in the p set
    - I think the delta is per-image in q, and is just aggregated over all images in p, but I'm not fully confident of this, as a result of notational confusion. It could also be one delta applied for all all images in q.
- Calculate the loss that results when you run forward the adversarially generated q against the normal p
- Train a combined loss that is a weighted combination of the normal p/q contrastive part and the adversarial p/q contrastive part

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

The authors show a small but relatively consistent improvement to performance using their method. Notably, this improvement is much stronger when using larger encoders (presumably because they have more capacity to learn from harder examples). One frustration I have with the empirics of the paper is that, at least in the main paper, they don't discuss the increase in training time required to calculate these perturbations, which, a priori, I would imagine to be nontrivial.

This paper argues that, in semi-supervised learning, it's suboptimal to use the same weight for all examples (as happens implicitly, when the unsupervised component of the loss for each example is just added together directly. Instead, it tries to learn weights for each specific data example, through a meta-learning-esque process. 

The form of semi-supervised learning being discussed here is label-based consistency loss, where a labeled image is augmented and run through the current version of the model, and the model is optimized to try to induce the same loss for the augmented image as the unaugmented one. The premise of the authors argument for learning per-example weights is that, ideally, you would enforce consistency loss less on examples where a model was unconfident in its label prediction for an unlabeled example. 

As a way to solve this, the authors suggest learning a vector of parameters - one for each example in the dataset - where element i in the vector is a weight for element i of the dataset, in the summed-up unsupervised loss. They do this via a two-step process, where first they optimize the parameters of the network given the example weights, and then the optimize the example weights themselves. To optimize example weights, they calculate a gradient of those weights on the post-training validation loss, which requires backpropogating through the optimization process (to determine how different weights might have produced a different gradient, which might in turn have produced better validation loss). This requires calculating the inverse Hessian (second derivative matrix of the loss), which is, generally speaking, a quite costly operation for huge-parameter nets. To lessen this cost, they pretend that only the final layer of weights in the network are being optimized, and so only calculate the Hessian with respect to those weights. They also try to minimize cost by only updating the example weights for the examples that were used during the previous update step, since, presumably those were the only ones we have enough information to upweight or downweight. With this model, the authors achieve modest improvements - performance comparable to or within-error-bounds better than the current state of the art, FixMatch. 

Overall, I find this paper a little baffling. It's just a crazy amount of effort to throw into something that is a minor improvement. A few issues I have with the approach: 

- They don't seem to have benchmarked against the simpler baseline of some inverse of using Dropout-estimated uncertainty as the weight on examples, which would, presumably, more directly capture the property of "is my model unsure of its prediction on this unlabeled example"
- If the presumed need for this is the lack of certainty of the model, that's a non-stationary problem that's going to change throughout the course of training, and so I'd worry that you're basically taking steps in the direction of a moving target
- Despite using techniques rooted in meta-learning, it doesn't seem like this models learns anything generalizable - it's learning index-based weights on specific examples, which doesn't give it anything useful it can do with some new data point it finds that it wasn't specifically trained on

Given that, I think I'd need to see a much stronger case for dramatic performance benefits for something like this to seem like it was worth the increase in complexity (not to mention computation, even with the optimized Hessian scheme)

This paper builds on top of a bunch of existing ideas for building neural conversational agents so as to control against generic and repetitive responses. 

Their model is the sequence-to-sequence model with attention (Bahdanau et al.), first trained with the usual MLE loss and fine-tuned with policy gradients to optimize for specific conversational properties. Specifically, they define 3 rewards:

1. Ease of answering — Measured as the likelihood of responding to a query with a list of hand-picked dull responses (more negative log likelihood is higher reward). 
2. Information flow — Consecutive responses from the same agent (person) should have different information, measured as negative of log cosine distance (more negative is better).
3. Semantic coherence — Mutual information between source and target (the response should make sense wrt query). $P(a|q) + P(q|a)$ where a is answer, q is question.

The model is pre-trained with the usual supervised objective function, taking source as concatenation of two previous utterances. Then they have two stages of policy gradient training, first with just a mutual information reward and then with a combination of all three. The policy network (sequence-to-sequence model) produces a probability distribution over actions (responses) given state (previous utterances). To estimate the gradient in an iteration, the network is frozen and responses are sampled from the model, the rewards for which are then averaged and gradients are computed for first L tokens of response using MLE and remaining T-L tokens with policy gradients, with L being gradually annealed to zero (moving towards just the long-term reward).

Evaluation is done based on length of dialogue, diversity (distinct unigram, bigrams) and human studies on

1. Which of two outputs has better quality (single turn)
2. Which of two outputs is easier to respond to, and
3. Which of two conversations have better quality (multi turn).

## Strengths

- Interesting results
    - Avoids generic responses
    - 'Ease of responding' reward encourages responses to be question-like
- Adding in hand-engineereed approximate reward functions based on conversational properties and using those to fine-tune a pre-trained network using policy gradients is neat.
- Policy gradient training also encourages two dialogue agents to interact with each other and explore the complete action space (space of responses), which seems desirable to identify modes of the distribution and not converge on a single, high-scoring, generic response.

## Weaknesses / Notes

- Evaluating conversational agents is hard. BLEU / perplexity are intentionally avoided as they don't necessarily reward desirable conversational properties.

This paper proposes a neural architecture that allows to backpropagate gradients though a procedure that can go through a variable and adaptive number of iterations. These "iterations" for instance could be the number of times computations are passed through the same recurrent layer (connected to the same input) before producing an output, which is the case considered in this paper. 

This is essentially achieved by pooling the recurrent states and respective outputs computed by each iteration. The pooling mechanism is essentially the same as that used in the really cool Neural Stack architecture of Edward Grefenstette, Karl Moritz Hermann, Mustafa Suleyman and Phil Blunsom \cite{conf/nips/GrefenstetteHSB15}. It relies on the introduction of halting units, which are sigmoidal units computed at each iteration and which gives a soft weight on whether the computation should stop at the current iteration.

Crucially, the paper introduces a new ponder cost $P(x)$, which is a regularization cost that penalizes what is meant to be a smooth upper bound on the number of iterations $N(t)$ (more on that below).

The paper presents experiment on RNNs applied on sequences where, at each time step t (not to be confused with what I'm calling computation iterations, which are indexed by n) in the sequence the RNN can produce a variable number $N(t)$ of intermediate states and outputs. These are the states and outputs that are pooled, to produce a single recurrent state and output for the time step t. During each of the $N(t)$ iterations at time step t, the intermediate states are connected to the same time-step-t input. After the $N(t)$ iterations, the RNN pools the $N(t)$ intermediate states and outputs, and then moves to the next time step $t+1$. To mark the transitions between time steps, an extra binary input is appended, which is 1 only for the first intermediate computation iteration. 

Results are presented on a variety of synthetic problems and a character prediction problem.