A Trichotomy for Transductive Online Learning

We present new upper and lower bounds on the number of learner mistakes in the `transductive' online learning setting of Ben-David, Kushilevitz and Mansour (1997). This setting is similar to standard online learning, except that the adversary fixes a sequence of instances $x_1,\dots,x_n$ to be labeled at the start of the game, and this sequence is known to the learner. Qualitatively, we prove a trichotomy, stating that the minimal number of mistakes made by the learner as $n$ grows can take only one of precisely three possible values: $n$, $Θ\left(\log (n)\right)$, or $Θ(1)$. Furthermore, this behavior is determined by a combination of the VC dimension and the Littlestone dimension. Quantitatively, we show a variety of bounds relating the number of mistakes to well-known combinatorial dimensions. In particular, we improve the known lower bound on the constant in the $Θ(1)$ case from $Ω\left(\sqrt{\log(d)}\right)$ to $Ω(\log(d))$ where $d$ is the Littlestone dimension. Finally, we extend our results to cover multiclass classification and the agnostic setting.

Paper

References (33)

Scroll for more · 21 remaining

Similar papers

Peer review

Reviewer Hibd7/10 · confidence 4/52023-06-25

Summary

This paper studies realizable transductive online learning on a fixed and *known* sequence $x_1,\dots,x_n$ (meaning the order is given). The authors state a trichotomy of error rates depending on the finiteness of the VC dimension and Littlestone dimension of the given hypothesis space. Along the way they improve a lower bound on the number of mistakes in terms of the Littlestone dimension. They use the relationship to the threshold dimension to achieve their bounds. Finally, they extend their results to the multi-class case.

Strengths

Nice overview. Well written. Easy to follow proofs. ---- rebuttal comment ----- As authors addressed multiple concerns and added additional results on the agnostic case, I raised my score from 6 to 7. I think this paper is valuable for the online learning community.

Weaknesses

Most of the claimed results are already known or exist implicitly. In particular, all the bounds used in the trichotomy are standard (either from Littlestone's Halving / SOA, or from the online vs offline paper). Even the additional $\log(n)$ lower bound given by the threshold dimension was essentially covered by an example (called "$\sigma_{\text{worst}}$") in [BKM97] (without using the threshold dimension explicitly though). The actual main novelty (which is still a decent and interesting contribution) is the improvement of the lower bound from $\sqrt{\log(LD(H))}$ to $\log(LD(H))$. The informal version of Thm 4.1. is slightly misleading. E.g., case 2 If $VC(H)=n$ the rate is still $\Theta(n)$ instead of $\log(n)$. I would suggest writing up the bounds for a finite instance space $|\mathcal{X}|=n$ as corollaries. This might be interesting for various settings (e.g., node classfication) and guide the intuition of the reader. I am not sure referring to the setting here with the sequence $(x_1,\dots,x_n)$ known in advance to the learner as "transductive" is the best idea. Transductive typically refers to the set (without knowing the order) $\\{x_1,\dots,x_n\\}$ to be known. Even one of the referenced papers [KK05], which the authors use to justify the name change from $M_{worst}$ to transductive, is merely referring the fact that the set is known but not necessarily the sequence (e.g., "the set of unlabeled examples is then presented to the learner") as transductive. Similarly for instance, the paper [CS13] refers to the fact that the set is known (instead of the sequence) as transductive "In this model, we have an arbitrary sequence of labeled examples $(x_1, y_1), . . . , (x_T , y_T)$, where only the set $\\{x_1, . . . , x_T\\}$ is known". I might be wrong though. Please discuss. Small typos and writing * line 126: $u_{\leq k}$ should probably be $x$? * line 202: "Claim claim" * line 235: "makes $d$ mistakes --> "makes at most $d$ mistakes". * line 236: What is $m$? Probably $d$? Given these weaknesses I would say that this work is rather incremental and has limited novelty. Nevertheless it offers some nice new connections (e.g., threshold dimension) and gives a nice overview (the trichotomy) about bounds which (implicitly or explicitly) exist in previous work.

Questions

Essentially the results (as most such online learning papers) leave the gap between $VC+\log n$ and $VC \cdot \log n$ open. Do you have any ideas how to close this gap? Sure you state that both bounds are tight on specific instance sequences, but maybe there is a different combinatorial parameter (probably depending on the particular sequence?) interpolating between "$+$" and "$\cdot$" allowing to get instance-dependent tight bounds. You briefly discuss some issues in the multi-class case with $|\mathcal{Y}|=\infty$. However there is the recent work by Brukhim et al [FOCS 2022] using the Daniely-Shalev-Shwartz (DS) dimension to characterize multi-class learnability even if $|\mathcal{Y}|=\infty$. Do these results apply in your setting by swapping the Natarajan dimension (or graph dimension etc.) with the DS dimension? Is the $\log(LD(H))$ lower bound best possible? I.e., are there specific examples where this is tight (meaning, same upper bound).

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

4 excellent

Presentation

3 good

Contribution

2 fair

Limitations

/

Reviewer GnLU5/10 · confidence 3/52023-07-06

Summary

This paper studies transductive online learning, which differs from standard learning in that the adversary must commit to a sequence of instances to be labelled by the learner at the start of the game. The adversary's strategy can thus only be adaptive w.r.t. the labeling of the sequence, not the sequence itself. The goal of the paper is to bound the number of mistakes made by learning algorithms on a sequence of length $n$. The main result is comprised of three different set ups: - If the VC dimension of a concept class is infinite, then the number of mistakes is $n$, - If the VC dimension is finite, but the Littlestone dimension is infinite, then the number of mistakes is logarithmic in $n$, - Finally, if the Littlestone dimension is finite, then the number of mistakes is constant (in the sense that it does not depend on $n$ -- it will however depend on the Littlestone dimension).

Strengths

- Conceptually interesting to have a trichotomy that depends on the VC and Littlestone dimensions - The paper is well-written and easy to follow, especially the proofs, which are clear and well-explained - The proof of Theorem 3.1 is interesting, and carries the technical weight of the paper, in addition to being a significant improvement on previous lower bounds - This paper seems of interest to the online learning theory community

Weaknesses

- Apart from the proof of Theorem 3.1, the proofs seem relatively straightforward, so the originality/novelty of the work is mainly conceptual, not technical. More specifically, items 1 and 3 from Theorem 4.1 are applications of the definitions of the VC and Littlestone dimensions -- but Theorem 4.1 is the main contribution of the paper, apart from the lower bound. Item 2 on appearance seems more substantial but l.218-225 are simply explaining the Halving algorithm, leaving the main technical tool to be an application of the Sauer-Shelah-Perles lemma. - A substantial improvement would come from getting a tighter bound w.r.t. the Littlestone dimension for the constant in item 3 of Theorem 4.1, as for now the proof (l.234-236) follows directly from the definition of the Littlestone dimension (of course, easier said than done! I just think there is some technical weight missing in the paper) I was hesitating between a 4 and 5, but opted for 5 because of Thm 3.1 and the conceptual contribution.

Questions

Questions - l.236: what is $m$? Do you mean $d$? - Is there a fundamental difference in the derivation of the results for the multiclass setting (vs binary)? Comments: - Informal version of Thm 4.1 (l.69): I think it should be specified that the constants in cases 2 and 3 hide a dependence on the VC and Littlestone dimensions, respectively. - l.202: "Claim claim" - Perhaps include a future work/conclusion section?

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

4 excellent

Contribution

2 fair

Limitations

N/A

Reviewer e1Y36/10 · confidence 3/52023-07-09

Summary

The paper considers the realizable case in the transductive online learning setting. It shows that, for a sequence of length $n$, the number of errors is $\Theta(n)$, $\Theta(\log n)$ or $\Theta(1)$ depending on the finiteness of the VC dimension and the Littlestone dimension. In the last case, the paper also improves the dependence on the Littlestone dimension $D_L$ from $\Omega(\sqrt{D_L})$ to $\Omega(D_L)$.

Strengths

The paper looks sound and their main result is clear.

Weaknesses

I have the following concerns: 1. The significance of the main contribution. In Theorem 4.1, the only non-trivial case is the upper bound in Item 2, and, as I understand, it follows from [KK05]. Maybe pointing out the trichotomy is significant in itself, but I'm not in the area, so I can't judge it. 2. Presentation should be improved. It took me some time to parse Definition 2.4, and I could do this only because I knew the definition before. Both this definition and the reasoning in lines 185-188 are better replaced with pictures. 3. I didn't judge by this point, but I suspect that not all relevant references are included. E.g. I think that some papers [KK05] should potentially be cited. Minor issues: * Line 202: "claim" is repeated * Why did you move some material to the supplementary? There should be enough space in the main body.

Questions

* Could the multi-dimensional case be stated in terms of graph dimension? Then there would be no $\log k$ factor.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

2 fair

Contribution

1 poor

Limitations

The paper only considers the realizable case

Reviewer gZ3U5/10 · confidence 4/52023-07-14

Summary

This work considers transductive online learning where the pool of examples to be labeled is fixed and known to the learner. Thus, the adversary can control the order that the points are presented to the learner and the label that the learner receives in each round. The goal of the learner is to predict labels in each round so that they do as few mistakes as possible. In this work the authors provide new results on the minimum possible number of mistakes. They show that the minimal mistakes can either be constant, grow logarithmically, or grow linearly with the number of examples $n$ that are to be labeled by the learner. More precisely, they show that if the VC dimension of the class is infinite then the number of mistakes is $n$, if the VC dimension is bounded but the Littlestone dimension is infinite then the number of mistakes is $\Theta(\log n)$, and when the Littlestone dimension is bounded the number of mistakes is constant. Morever, when the Littlestone dimension is at most $d$ the authors show a $\Omega(\log d)$ lower bound on the number of mistakes.

Strengths

The transductive learning model is a well-motivated, and interesting learning model that may often be more realistic than the fully adversarial (standard) online setting where the adversary can choose the examples to be labeled. This work provides a new trichotomy result and an improved bound ($\Omega(\log d)$ instead of $\Omega(\sqrt{\log d})$ that was shown in the prior work) on the number of mistakes under bounded Littlestone dimension. Moreover, the argument for the $\Omega(\log d)$ is nice.

Weaknesses

My main concern about the main result (the trichotomy of Theorem 4.1) of this work follows rather easily from prior works. It is not hard to show that when the VC dimension is infinite the adversary can always make the learner to do $n$ mistakes. Moreover, with bounded VC dimension (less than or equal to $d$), the algorithm that achieves the $d \log (n/d)$ mistake bound is the standard halving algorithm that has been used extensively in the online learning literature. Finally for the third assertion of Theorem 4.1, when Littlestone dimension is bounded the work of Littlestone '87 directly yields a finite mistake bound. The multiclass generalization of the trichotomy is a nice generalization but follows in the same way as the binary result and does not seem to add much to the paper from a technical point of view.

Questions

See weaknesses. Stating the key ideas or differences of the techniques of this work compared to the prior works could help.

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

4 excellent

Presentation

4 excellent

Contribution

3 good

Limitations

Yes.

Reviewer a8Lo7/10 · confidence 4/52023-07-23

Summary

This paper studies transductive online learning. In this setup, the adversary fixes the sequence of unlabeled instances and reveals labels sequentially following learner's prediction in each round. The paper shows a trichotomy of rates for binary classification: $n$, $\Theta(\log{n})$ and $\Theta(1)$ based on finiteness/infiniteness of VC and Littlestone dimension of the hypotheses class. The authors also extend the result to multiclass setting establishing a similar trichotomy depending on Natarajan and multiclass extension of Littlestone dimension. Finally, the authors provide a quadratic improvement in the known lower bound in the case $\Theta(1)$.

Strengths

1. The paper identifies the trichotomy of rates in online transductive learning. Such trichotomy has been identified in the recently introduced setting of universal learning [BHMvHY2021]. As far as I know, this is the first paper to note such trichotomy in any variant of online learning model. Qualitatively, the paper shows that VC, instead of more restrictive Littlestone, is necessary and sufficient for learnability of the hypotheses class in transductive online setting. 2. The paper studies a theoretically interesting problem and motivates the problem through a meaningful example. Apart from the proof in Section 6 which I found to be written sloppily, the paper is well-written and easy to follow. 3. The proof technique of Theorem 3.1 is novel and differs from standard lower bound techniques in online learning literature. [BHMvHY2021] Olivier Bousquet, Steve Hanneke, Shay Moran, Ramon van Handel, and Amir Yehudayoff. 2021. A theory of universal learning. In Proceedings of the 53rd Annual ACM SIGACT Symposium on Theory of Computing (STOC 2021). Association for Computing Machinery, New York, NY, USA, 532–541. https://doi.org/10.1145/3406325.3451087

Weaknesses

My general impression of the paper is that there is a room for a more comprehensive treatment of the material considered here. 1. I might be speaking with a hindsight bias here, but I think most of proofs regarding trichotomy (Theorem 4.1) is straightforward except that of Claim 3.4 (also see my question below). In view of that, I would have preferred to see a sharper analysis of mistake bounds in Items 2 and 3 . Namely: (a) What is the optimal dependence of $\text{VC}(\mathcal{H})$ and $n$ for the case $\text{VC}(\mathcal{H}) < \infty$ and $\text{LD} (\mathcal{H}) = \infty$? (b) Is there an algorithm that exploits the transductive nature of the game to get a mistake bound better than $\text{LD}(\mathcal{H})$ when it is finite? 2. The exposition of proof of Theorem 3.1, arguably the main technical contributions of the paper, can be improved. For instance, what is $j$ in $\varepsilon_t = \varepsilon^{(j+1)}$ in line 267? What is the role of the constant $C_1$ in the proof/result? Line 263 claims we can take $C_1=5$? If so, why not just use $5$ for a clearer presentation (although I dont see why it cant be 2)? What is superscript $1$ in $C_1$ indexing? The paragraph following line 270 (that analyzes the adversary) can also be improved in its clarity. 3. The authors could have considered agnostic setting as well. What is the characterization of learnability for online agnostic transductive learning? Does one observe such trichotomy of rates?

Questions

1. Does Theorem 3.1 imply the $\log{n}$ lower bound in Item (2) of Theorem 4.1 using the fact that $\text{LD}(\mathcal{H}) \geq n$ for each $n$? If so, is the discussion of Threshold dimension and Claim 3.4 required to establish the trichotomy for binary case? What about multiclass extension? 2. In Line 44-45, the authors claim that neither party benefit from using randomness. Do authors mean that there wont be any quantitative difference (up to a constant factor) between deterministic and randomized learning rules? This is most likely true but I would like authors to justify their statement instead of just claiming it.

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Not applicable.

Reviewer GnLU2023-08-11

Response

Thanks for the general reply and for the specific answers to my questions. I think it would be worth highlighting the technical contribution of the multi-class setting in the main paper, even at a high level -- otherwise it seems like there isn't much difference with the binary setting. Given the page limit, there is definitely enough room to add this discussion/overview!

Reviewer Hibd2023-08-14

Dear reviewer GnLU, dear authors, I'd really like to see more comments on the agnostic case. Do the authors have first results? In standard online learning (without the sequence known in advance to the learner) tight results exist (see e.g., [1,2] and perhaps some more). Do these apply here? Best, Hibd [1] Ben-David, Shai, Dávid Pál, and Shai Shalev-Shwartz. "Agnostic Online Learning." COLT 2009 [2] Hanneke, Steve, et al. "Multiclass online learning and uniform convergence. COLT 2023

Authorsrebuttal2023-08-16

Thank you for your interest! We have added a detailed discussion, please see **_General Answer 2: The Agnostic Case_** above.

Authorsrebuttal2023-08-18

Also, following your comment, we will be sure to highlight the technical contribution of the multi-class setting in the final version of the paper :)

Reviewer a8Lo2023-08-11

Thank you for your general reply and for answering my specific question. I look forward to reading the polished version of the proof, and I think it would be helpful to include more discussion on proof techniques for multiclass settings in the main text. Given the response by the authors, I am happy to raise the score to 5.

Authorsrebuttal2023-08-16

We are thrilled to hear that you are raising your score!! Per your question on the agnostic case: **The authors could have considered agnostic setting as well. What is the characterization of learnability for online agnostic transductive learning? Does one observe such trichotomy of rates?** Please see our **_General Answer 2: The Agnostic Case_** above. We feel this is a valuable addition!

Reviewer a8Lo2023-08-16

Reply to authors

Thank you for providing details about the agnostic case. Although the lower-bound proof uses a standard technique form [1] and the upperbound proof involves a standard Sauer-Selah-Pereles argument, I think the result makes the overall story complete. Also, I do not think proof techniques being standard is a weakness of the paper, but rather the strength of our tools. Here is my revised assessment of the paper after the response by the authors: 1. Identifying the trichotomy was already an important contribution of this paper. 1. The qualitative characterization of learnability in this setting is interesting and overall complete. Just to reiterate, a binary hypothesis class $\mathcal{H}$ is learnable in a transductive setting if and only if $ \text{VC}(\mathcal{H}) < \infty$. And the rates involving $\text{VC}(\mathcal{H})$ are very reasonable. As we all know, removing that $\sqrt{\log{T}}$ is typically difficult in the online setting. 2. Correct me if I am wrong, but your argument for the agnostic case should easily extend to a multiclass setting with finite $|\mathcal{Y}|$ as well, right? One can consider a Natarajan shattered set and sample uniformly between labels of two witness functions for each x for the lower bound, and use multiclass Sauer-Selah-Pereles for the upper bound. 3. Next, an interesting question is what happens in the case where $|\mathcal{Y}|$ is unbounded? Your example in the paper showing that the analogy breaks down is interesting. However, what I find more interesting is your answer to reviewer Hibd, which shows that the DS dimension does not characterize learnability in the transductive setting when $|\mathcal{Y}| $ is unbounded. Finding a dimension that characterizes learnability in the transductive setting when $|\mathcal{Y}| $ is unbounded is an exciting open problem. I raised my score from 4 to 5 because the authors addressed my concerns regarding the necessity of threshold dimension and the proof of the lower bound of $\log{n}$. Given that they also worked out the details for the agnostic case (one of the points of weakness in my original assessment)and provided a very interesting example (I would like authors to include this example in the paper) showing DS dimension does not characterize learnability makes this paper complete and strong in my opinion. Following these modifications and addition, I think this will be an important contribution to the field and should be accepted. In view of my revised assessment, I am raising my score to 7 and my confidence back to 4. Also, I am willing to strongly vouch for the acceptance of this paper. [1] Shai Ben-David, Dávid Pál, and Shai Shalev-Shwartz. Agnostic Online Learning. COLT 2009.

Authorsrebuttal2023-08-18

That is wonderful! We are thrilled to hear this! :) We will be sure to include the example regarding the DS dimension in the final version. I believe you are correct about extending the agnostic arguments to the case of finite $k = |\mathcal Y|$ and we will include that in the final version as well!

Reviewer e1Y32023-08-13

Thank you for the reply. Can you please answer the question about the graph dimension?

Authorsrebuttal2023-08-14

Thank you for calling this to our attention! Apologies, we accidentally omitted our answer to the question about the graph dimension. We actually don't see a reason why using the graph dimension would allow us to eliminate the $\log(k)$ factor from the upper bound in Theorem B.2. Our upper bound uses the Sauer–Shelah–Perles lemma for the Natarajan dimension, which states that a class with Natarajan dimension $d$ and $k$ possible labels has at most $\left(e n k^2/d\right)^d$ functions, so the halving algorithm will make at most $O(d \log (n k / d))$ mistakes. Similarly, note that a class with graph dimension $d$ can also have a number of functions that grows roughly like $k^d$, so the same upper bound would follow if we apply the same proof strategy to the graph dimension. For example, the class of all functions $f: [d] \to [k]$ has graph dimension $d$ and contains $k^d$ functions. While this shows that the same proof would not yield an improved upper bound, it is of course entirely possible that a different proof might do so.

Authorsrebuttal2023-08-18

Dear Reviewer e1Y3, In your review, you mentioned the following points: * **"The significance of the main contribution. In Theorem 4.1, the only non-trivial case is the upper bound in Item 2, and, as I understand, it follows from [KK05]."** We feel this is not accurate. Please see **[General Answer 1](https://openreview.net/forum?id=iSd8g75QvP&noteId=wuoq32YtUM)**. * **"Could the multi-dimensional case be stated in terms of graph dimension?"** We have answered that question. * **You commented on related works.** In the final version, we will expand our discussion of related works, including the paper [KK05] that you mentioned. * **"The paper only considers the realizable case."** Following your comment, we have added a fairly complete analysis of the agnostic case. Please see **[General Answer 2: The Agnostic Case](https://openreview.net/forum?id=iSd8g75QvP&noteId=NpnLZNlj4s)**. Thus, we feel that we have fully addressed all the concerns raised in your review. Additionally, there has been some significant discussion with the other reviewers. In particular, we invite you to read **[this very nice summary by Reviewer A8Lo](https://openreview.net/forum?id=iSd8g75QvP&noteId=oTP7gwPBR2)**. At this point, do you have any major concerns about our paper? If not, would you be willing to reconsider your score for our paper?

Reviewer e1Y32023-08-18

Thank you for the replies. I agree that the agnostic case makes the paper stronger. After all the clarifications, I do believe that the paper has enough technical weight, and I raised my score.

Reviewer Hibd2023-08-14

Thank you for your clarifications. Interesting that the DS dimension does not characterise this setting here! I like the terms set-transductive vs sequence-transductive. Either way, some paragraph on related work discussing this ambiguity would be more than sufficient and helpful for readers. Do you think that your derivation of the $log(TD(H))$ lower bound is essentially using the same arguments as the one discussed for the specific sequence $\sigma_\text{worst}$ in BKM97? Please comment. Another small comment. Maybe indicate in Thm 4.1 that $C(H)\leq d$ (with $d=Ldim$) otherwise it is unclear how large this quantity might be without looking at the proof. Finally, I'd be happy if you can give more comments on the agnostic setting mentioned by another reviewer. Do you have some first results?

Authorsrebuttal2023-08-16

Thank you for your interest! We have added a detailed discussion on the agnostic case, please see **_General Answer 2: The Agnostic Case_** above. We will address your remaining points soon :)

Reviewer Hibd2023-08-16

Thank you I raised my score to a 7. And yes with online node classification / online learning on graphs I've mean papers like * N. Cesa-Bianchi, C. Gentile, F. Vitale, G. Zappella Random Spanning Trees and the Prediction of Weighted Graphs (JMLR 2013) * Mark Herbster, Stephen Pasteris, Shaona Ghosh "Online prediction at the limit of zero temperature" (NeurIPS 2015) and the one you mentioned (and other related papers)

Authorsrebuttal2023-08-16

The Agnostic Case

**General Answer 2: The Agnostic Case** A number of reviewers have asked for our thoughts concerning the agnostic case. Let us briefly review what is known, and summarize our initial observations: In the _standard_ (non-transductive) agnostic online setting, Ben-David et al. [1] showed that $R_T(\mathcal{H})$, the agnostic online regret for a hypothesis class $\mathcal{H}$ with Littlestone dimension $d_L$ for horizon size $T$ satisfies $\Omega(\sqrt{d_L \cdot T}) \leq R_T(\mathcal{H}) \leq O(\sqrt{d_L \cdot T \log T}).$ Later, Alon et al. [2] showed that the precise regret is $R_T(\mathcal{H}) = \Theta(\sqrt{d_L \cdot T})$. For the _transductive_ agnostic online setting, we make the following novel observation: **Claim.** _Let $\mathcal{H}$ be a hypothesis class with VC dimension $d_V > 0$. Then the agnostic transductive regret for $\mathcal{H}$ with horizon $T$ is $R_T(\mathcal{H}) = \tilde{\Theta}(\sqrt{d_V\cdot T})$._ An upper bound of $O(\sqrt{d_{V}\cdot T\log(T/d_{V})})$ follows from the Perles–Sauer–Shelah lemma + the multiplicative weights algorithm. **Proof sketch for lower bound.** Let $d=d_{V}$, and let $x_1,\dots,x_d$ be VC shattered by $\mathcal{H}$. Assume for simplicity that $T = kd$ for some $k \in \mathbb{N}$. The transductive adversary selects the sequence $x_1^1,\dots,x_1^k,x_2^1,\dots,x_2^k,\dots,x_d^1,\dots,x_d^k$, where $x_i^j = x_i$ for all $j \in [k]$ and $i \in [d]$. The adversary chooses all labels to be i.i.d. uniform random bits. By Khinchine’s Inequality (See Lemma 8.2 in [3]), for each $i \in [d]$ the expected gap between the majority label and the minority label in the subsequence $x_i^1,\dots,x_i^k$ is of order $\Omega(\sqrt{k})$. Hence, by an averaging argument, any learner will have excess error of order $\Omega(\sqrt{k})$ on roughly half of the $d$ subsequences $x_i^1,\dots,x_i^k$. This implies the desired lower bound of $\Omega(\sqrt{k} \cdot d) = \Omega(\sqrt{d\cdot T})$. Additionally: * For trivial classes (e.g., a class that contains a single function), the regret is $0$. * For classes with finite VC and finite Littlestone dimension, the regret is $\Theta(\sqrt{d_L \cdot T})$ (by [2]), namely, the $\log(T)$ factor from the upper bound for VC classes is removed. * For classes with infinite VC, the regret is $T$. Hence, for non-trivial classes, finiteness of the VC dimension determines the asymptotic dependence of the regret on $T$ up to logarithmic factors. However, the regret for VC classes remains open because there is a gap of a $\sqrt{\log(T)}$ multiplicative factor between the best known upper and lower bounds. In other words, we don’t know whether there is just one case or several cases. _____ [1] Shai Ben-David, Dávid Pál, and Shai Shalev-Shwartz. Agnostic Online Learning. COLT 2009. [2] Alon, N., Ben-Eliezer, O., Dagan, Y., Moran, S., Naor, M. and Yogev, E. Adversarial laws of large numbers and optimal regret in online classification. STOC 2021. [3] N. Cesa-Bianchi and G. Lugosi. Prediction, learning, and games. Cambridge University Press, 2006.

Authorsrebuttal2023-08-16

Please see our **_General Answer 2: The Agnostic Case_** above. We feel this is a nice addition!

Reviewer gZ3U2023-08-17

Thanks for the Response

I would like to thank the authors for their response. I agree that the agnostic case is a nice addition which makes the story better. I am somewhat skeptical about adding new results post submission but I am willing to vote for acceptance (I will raise my score to 5).

Authorsrebuttal2023-08-18

Briefly addressing remaining comments: **1. Do you think that your derivation of the lower bound is essentially using the same arguments as the one discussed for the specific sequence $\sigma_{\text{worst}}$ in BKM97? Please comment.** Yes, Claim 3.4 in our paper generalizes the example from BKM97. Edit: To clarify, that part of our argument is similar to that example. However, our proof of the $\log(n)$ lower bound in the trichotomy also uses the threshold dimension and the connection between threshold dimension and Ldim, which are novel in this context. **2. Maybe indicate in Thm 4.1 that $C(H) < d$ (with $d = Ldim$) otherwise it is unclear how large this quantity might be without looking at the proof.** Yes, this is a good suggestion. We will add this in the final version. ___ Thank you for your time and careful reading of our paper!

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC