Summary: This paper proposes a stochastic online learning method for the task of structured prediction. In this setting, the learner doest not get the correct structured output during training. Instead, it only gets bandit feedback from the labeler. The paper first proposes an online learning algorithm that learns model parameters via stochastic gradient descent; generalizes the learning method to pair-wise comparison of structured outputs; provides an optimization approach with Cross-Entropy Minimization; and theoretically analyzes the convergence property of the optimization approach. Pros: The paper proposes an online stochastic learning algorithm for minimizing the expected loss of structured predictions; gives a method of learning from pair-wise comparisons; and theoretical analyze the convergence rate.

Most of the results established in the paper would, in the special case of binary classification, trivially follow from the known upper and lower bounds on sample complexity based on the VC dimension. However, the results were not previously known for multiclass learning, and other general loss functions. The results for the 0-1 loss are not particularly surprising, but it is good to know that, for instance, in multiclass classification with the 0-1 loss, the complexity measure in the agnostic sample complexity is the same as that in the realizable-case (up to log factors, but no extra factors such as log( Y) not present in the realizable-case sample complexity). They also prove a tighter lower bound than previously known for the sample complexity of uniform convergence for multiclass classification in Theorem 3.6. The techniques used in the proofs are mostly straightforward or have appeared in other related contexts previously.

This paper is interesting and deals with new kind of results introducing computational aspects in standard minimax theory. The phenomenon illustrated is new to me, and present some limitation of the computationally tractable algorithm w.r.t. "theoretical" ones that could be considered in the classical minimax theory. However, due to the relative novelty of the framework, it would be important that basic definitions and properties be better presented. In the following there is only one model investigated.

The paper is well written and motivated. In particular the problem considered is relevant. On the downside there are some issues related to the interpretability of the presented results: - In Theorem 1 the generalization error is bounded in terms of the additive or multiplicative empirical margin losses. However their formulation at Eq. (5) and (6) is hard to interpret and would benefit from a comment. This is problematic since it is not clear how these quantities are related to the algorithmic approaches discussed in Sec. 5.

The idea of using tensor networks for machine learning problems is interesting. Unfortunately, there are many important issues that prevent the publication of the manuscript as it is now. My main concerns are the following: 1- Some parts of the paper are hard to read. The paper applies and adapts some idea about tensor networks that are used/developed by the Physics community to ML problems. The authors have done a good job in illustrating the idea and the intuition by many figures.

This work proposes to use semi-supervised learning, in the form of an unsupervised loss term, for improving the regularization capacity of CNNs. The idea (and the proposed loss) is conceptually simple and enforces stability explicitly by minimizing the difference between predictions corresponding to the same input data point. The paper focuses mainly on the experimental side, devoting the largest part in presenting results when adding the new loss on standard supervised CNNs. This is the stronger aspect of this work, with the weaker being the lack (or the definition) of baselines and the lack of some form of theoretical justification, derivation or discussion. Novelty/originality: The main contribution is the application of the unsupervised loss term for controlling the stability of the predictions under transformations or stochastic variability.

The paper is a superbly written account of a simple idea that appears to work very well. The approach can straightforwardly be applied to existing max-likelihood (ML) trained models in order to in principle take into account the task reward during training and is computationally much more efficient than alternative non ML based approaches. This work risks being underappreciated as proposing but a simple addition of artificial structured-label noise, but I think the specific link with structured output task reward is sufficiently original, and the paper also uncovers important theoretical insight by revealing the formal relationship between the proposed reward augmented ML and RL-based regularized expected reward objectives. So while it works surprisingly well, you haven't yet clearly demonstrated empirically that using a truly *task-reward derived* payoff distribution is beneficial. One way to convincingly demonstrate that would be if you did your envisioned BLEU importance reweighted sampling, and were able to show that it improves the BLEU test score over your current simpler edit-distance based label noise.

Overall the paper presents a novel application of CNNs to co-evolution Protein Contact Prediction (PCP) with state-of-the-art results. The paper is a strong contribution to the field of coevolution PCP. The method has the potential to improve ab initio structure prediction accuracy, which could have wide ranging impacts on studying proteins without solved structural complexes. The paper itself contains flaws in the experimental setup, how the methods are described, and how the results are presented. Key issues: - Experimental setup: not enough unique training examples are used to compare to existing methods. More than 100K examples exist for this specific task, but only 307 are used.

Differentially private training offers a protection which is usually interpreted as a guarantee against membership inference attacks. By proxy, this guarantee extends to other threats like reconstruction attacks attempting to extract complete training examples. Recent works provide evidence that if one does not need to protect against membership attacks but instead only wants to protect against a training data reconstruction, then utility of private models can be improved because less noise is required to protect against these more ambitious attacks. We investigate this question further in the context of DP-SGD, a standard algorithm for private deep learning, and provide an upper bound on the success of any reconstruction attack against DP-SGD together with an attack that empirically matches the predictions of our bound. Together, these two results open the door to fine-grained investigations on how to set the privacy parameters of DP-SGD in practice to protect against reconstruction attacks.

This paper considers contrastive training for cross-modal 0-shot transfer wherein a pre-trained model in one modality is used for representation learning in another domain using pairwise data. The learnt models in the latter domain can then be used for a diverse set of tasks in a 0-shot way, similar to Contrastive Language-Image Pre-training (CLIP) and Locked-image Tuning (LiT) that have recently gained considerable attention. Classical contrastive training employs sets of positive and negative examples to align similar and repel dissimilar training data samples. However, similarity amongst training examples has a more continuous nature, thus calling for a more non-binary' treatment. To address this, we propose a new contrastive loss function called Continuously Weighted Contrastive Loss (CWCL) that employs a continuous measure of similarity. With CWCL, we seek to transfer the structure of the embedding space from one modality to another.