Pub Date : 2022-11-16DOI: 10.48550/arXiv.2211.09101
Lunjia Hu, Charlotte Peale
In many learning theory problems, a central role is played by a hypothesis class: we might assume that the data is labeled according to a hypothesis in the class (usually referred to as the realizable setting), or we might evaluate the learned model by comparing it with the best hypothesis in the class (the agnostic setting). Taking a step beyond these classic setups that involve only a single hypothesis class, we introduce comparative learning as a combination of the realizable and agnostic settings in PAC learning: given two binary hypothesis classes $S$ and $B$, we assume that the data is labeled according to a hypothesis in the source class $S$ and require the learned model to achieve an accuracy comparable to the best hypothesis in the benchmark class $B$. Even when both $S$ and $B$ have infinite VC dimensions, comparative learning can still have a small sample complexity. We show that the sample complexity of comparative learning is characterized by the mutual VC dimension $mathsf{VC}(S,B)$ which we define to be the maximum size of a subset shattered by both $S$ and $B$. We also show a similar result in the online setting, where we give a regret characterization in terms of the mutual Littlestone dimension $mathsf{Ldim}(S,B)$. These results also hold for partial hypotheses. We additionally show that the insights necessary to characterize the sample complexity of comparative learning can be applied to characterize the sample complexity of realizable multiaccuracy and multicalibration using the mutual fat-shattering dimension, an analogue of the mutual VC dimension for real-valued hypotheses. This not only solves an open problem proposed by Hu, Peale, Reingold (2022), but also leads to independently interesting results extending classic ones about regression, boosting, and covering number to our two-hypothesis-class setting.
在许多学习理论问题中,假设类扮演着核心角色:我们可能会假设数据是根据类中的假设标记的(通常称为可实现设置),或者我们可能会通过将其与类中的最佳假设(不可知论设置)进行比较来评估学习模型。在这些只涉及单个假设类的经典设置之外,我们将比较学习作为PAC学习中可实现和不可知论设置的组合引入:给定两个二元假设类$S$和$B$,我们假设数据根据源类$S$中的假设进行标记,并要求学习模型达到与基准类$B$中的最佳假设相当的准确性。即使$S$和$B$都具有无限的VC维,比较学习仍然具有很小的样本复杂度。我们证明了比较学习的样本复杂性是由共同的VC维$mathsf{VC}(S,B)$表征的,我们将其定义为由$S$和$B$破碎的子集的最大大小。我们还在在线设置中显示了类似的结果,其中我们根据互Littlestone维$mathsf{Ldim}(S,B)$给出了遗憾表征。这些结果也适用于部分假设。我们还表明,表征比较学习的样本复杂性所必需的见解可以应用于表征可实现的多精度和多校准的样本复杂性,使用互散脂维(实值假设的互VC维的类似物)。这不仅解决了Hu, Peale, Reingold(2022)提出的一个开放问题,而且还将关于回归,提升和覆盖数的经典结果扩展到我们的双假设类设置中,从而产生了独立有趣的结果。
{"title":"Comparative Learning: A Sample Complexity Theory for Two Hypothesis Classes","authors":"Lunjia Hu, Charlotte Peale","doi":"10.48550/arXiv.2211.09101","DOIUrl":"https://doi.org/10.48550/arXiv.2211.09101","url":null,"abstract":"In many learning theory problems, a central role is played by a hypothesis class: we might assume that the data is labeled according to a hypothesis in the class (usually referred to as the realizable setting), or we might evaluate the learned model by comparing it with the best hypothesis in the class (the agnostic setting). Taking a step beyond these classic setups that involve only a single hypothesis class, we introduce comparative learning as a combination of the realizable and agnostic settings in PAC learning: given two binary hypothesis classes $S$ and $B$, we assume that the data is labeled according to a hypothesis in the source class $S$ and require the learned model to achieve an accuracy comparable to the best hypothesis in the benchmark class $B$. Even when both $S$ and $B$ have infinite VC dimensions, comparative learning can still have a small sample complexity. We show that the sample complexity of comparative learning is characterized by the mutual VC dimension $mathsf{VC}(S,B)$ which we define to be the maximum size of a subset shattered by both $S$ and $B$. We also show a similar result in the online setting, where we give a regret characterization in terms of the mutual Littlestone dimension $mathsf{Ldim}(S,B)$. These results also hold for partial hypotheses. We additionally show that the insights necessary to characterize the sample complexity of comparative learning can be applied to characterize the sample complexity of realizable multiaccuracy and multicalibration using the mutual fat-shattering dimension, an analogue of the mutual VC dimension for real-valued hypotheses. This not only solves an open problem proposed by Hu, Peale, Reingold (2022), but also leads to independently interesting results extending classic ones about regression, boosting, and covering number to our two-hypothesis-class setting.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"192 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114249718","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-07DOI: 10.48550/arXiv.2211.03396
Sourav Chakraborty, Anna G'al, Sophie Laplante, Rajat Mittal, Anupa Sunny
We introduce and study Certificate Game complexity, a measure of complexity based on the probability of winning a game where two players are given inputs with different function values and are asked to output $i$ such that $x_ineq y_i$ (zero-communication setting). We give upper and lower bounds for private coin, public coin, shared entanglement and non-signaling strategies, and give some separations. We show that complexity in the public coin model is upper bounded by Randomized query and Certificate complexity. On the other hand, it is lower bounded by fractional and randomized certificate complexity, making it a good candidate to prove strong lower bounds on randomized query complexity. Complexity in the private coin model is bounded from below by zero-error randomized query complexity. The quantum measure highlights an interesting and surprising difference between classical and quantum query models. Whereas the public coin certificate game complexity is bounded from above by randomized query complexity, the quantum certificate game complexity can be quadratically larger than quantum query complexity. We use non-signaling, a notion from quantum information, to give a lower bound of $n$ on the quantum certificate game complexity of the $OR$ function, whose quantum query complexity is $Theta(sqrt{n})$, then go on to show that this ``non-signaling bottleneck'' applies to all functions with high sensitivity, block sensitivity or fractional block sensitivity. We consider the single-bit version of certificate games (inputs of the two players have Hamming distance $1$). We prove that the single-bit version of certificate game complexity with shared randomness is equal to sensitivity up to constant factors, giving a new characterization of sensitivity. The single-bit version with private randomness is equal to $lambda^2$, where $lambda$ is the spectral sensitivity.
{"title":"Certificate games","authors":"Sourav Chakraborty, Anna G'al, Sophie Laplante, Rajat Mittal, Anupa Sunny","doi":"10.48550/arXiv.2211.03396","DOIUrl":"https://doi.org/10.48550/arXiv.2211.03396","url":null,"abstract":"We introduce and study Certificate Game complexity, a measure of complexity based on the probability of winning a game where two players are given inputs with different function values and are asked to output $i$ such that $x_ineq y_i$ (zero-communication setting). We give upper and lower bounds for private coin, public coin, shared entanglement and non-signaling strategies, and give some separations. We show that complexity in the public coin model is upper bounded by Randomized query and Certificate complexity. On the other hand, it is lower bounded by fractional and randomized certificate complexity, making it a good candidate to prove strong lower bounds on randomized query complexity. Complexity in the private coin model is bounded from below by zero-error randomized query complexity. The quantum measure highlights an interesting and surprising difference between classical and quantum query models. Whereas the public coin certificate game complexity is bounded from above by randomized query complexity, the quantum certificate game complexity can be quadratically larger than quantum query complexity. We use non-signaling, a notion from quantum information, to give a lower bound of $n$ on the quantum certificate game complexity of the $OR$ function, whose quantum query complexity is $Theta(sqrt{n})$, then go on to show that this ``non-signaling bottleneck'' applies to all functions with high sensitivity, block sensitivity or fractional block sensitivity. We consider the single-bit version of certificate games (inputs of the two players have Hamming distance $1$). We prove that the single-bit version of certificate game complexity with shared randomness is equal to sensitivity up to constant factors, giving a new characterization of sensitivity. The single-bit version with private randomness is equal to $lambda^2$, where $lambda$ is the spectral sensitivity.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125905633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We consider the classic online learning and stochastic multi-armed bandit (MAB) problems, when at each step, the online policy can probe and find out which of a small number ($k$) of choices has better reward (or loss) before making its choice. In this model, we derive algorithms whose regret bounds have exponentially better dependence on the time horizon compared to the classic regret bounds. In particular, we show that probing with $k=2$ suffices to achieve time-independent regret bounds for online linear and convex optimization. The same number of probes improve the regret bound of stochastic MAB with independent arms from $O(sqrt{nT})$ to $O(n^2 log T)$, where $n$ is the number of arms and $T$ is the horizon length. For stochastic MAB, we also consider a stronger model where a probe reveals the reward values of the probed arms, and show that in this case, $k=3$ probes suffice to achieve parameter-independent constant regret, $O(n^2)$. Such regret bounds cannot be achieved even with full feedback after the play, showcasing the power of limited ``advice'' via probing before making the play. We also present extensions to the setting where the hints can be imperfect, and to the case of stochastic MAB where the rewards of the arms can be correlated.
{"title":"Online Learning and Bandits with Queried Hints","authors":"Aditya Bhaskara, Sreenivas Gollapudi, Sungjin Im, Kostas Kollias, Kamesh Munagala","doi":"10.48550/arXiv.2211.02703","DOIUrl":"https://doi.org/10.48550/arXiv.2211.02703","url":null,"abstract":"We consider the classic online learning and stochastic multi-armed bandit (MAB) problems, when at each step, the online policy can probe and find out which of a small number ($k$) of choices has better reward (or loss) before making its choice. In this model, we derive algorithms whose regret bounds have exponentially better dependence on the time horizon compared to the classic regret bounds. In particular, we show that probing with $k=2$ suffices to achieve time-independent regret bounds for online linear and convex optimization. The same number of probes improve the regret bound of stochastic MAB with independent arms from $O(sqrt{nT})$ to $O(n^2 log T)$, where $n$ is the number of arms and $T$ is the horizon length. For stochastic MAB, we also consider a stronger model where a probe reveals the reward values of the probed arms, and show that in this case, $k=3$ probes suffice to achieve parameter-independent constant regret, $O(n^2)$. Such regret bounds cannot be achieved even with full feedback after the play, showcasing the power of limited ``advice'' via probing before making the play. We also present extensions to the setting where the hints can be imperfect, and to the case of stochastic MAB where the rewards of the arms can be correlated.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114026160","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-04DOI: 10.48550/arXiv.2211.02257
Guy Blanc, Caleb M. Koch, Jane Lange, Carmen Strassle, Li-Yang Tan
In the certification problem, the algorithm is given a function $f$ with certificate complexity $k$ and an input $x^star$, and the goal is to find a certificate of size $le text{poly}(k)$ for $f$'s value at $x^star$. This problem is in $mathsf{NP}^{mathsf{NP}}$, and assuming $mathsf{P} ne mathsf{NP}$, is not in $mathsf{P}$. Prior works, dating back to Valiant in 1984, have therefore sought to design efficient algorithms by imposing assumptions on $f$ such as monotonicity. Our first result is a $mathsf{BPP}^{mathsf{NP}}$ algorithm for the general problem. The key ingredient is a new notion of the balanced influence of variables, a natural variant of influence that corrects for the bias of the function. Balanced influences can be accurately estimated via uniform generation, and classic $mathsf{BPP}^{mathsf{NP}}$ algorithms are known for the latter task. We then consider certification with stricter instance-wise guarantees: for each $x^star$, find a certificate whose size scales with that of the smallest certificate for $x^star$. In sharp contrast with our first result, we show that this problem is $mathsf{NP}^{mathsf{NP}}$-hard even to approximate. We obtain an optimal inapproximability ratio, adding to a small handful of problems in the higher levels of the polynomial hierarchy for which optimal inapproximability is known. Our proof involves the novel use of bit-fixing dispersers for gap amplification.
在认证问题中,给算法一个证书复杂度为k的函数$f$和一个输入$x^star$,目标是为$f$在$x^star$处的值找到一个大小为$le text{poly}(k)$的证书。这个问题在$mathsf{NP}^{mathsf{NP}}$中,并且假设$mathsf{P} ne mathsf{NP}$不在$mathsf{P}$中。之前的工作,可以追溯到1984年的Valiant,因此寻求通过对$f$施加假设(如单调性)来设计有效的算法。我们的第一个结果是一般问题的$mathsf{BPP}^{mathsf{NP}}$算法。关键因素是变量平衡影响的新概念,这是一种纠正函数偏差的影响的自然变体。均衡影响可以通过均匀生成来准确估计,经典的$mathsf{BPP}^{mathsf{NP}}$算法用于后一项任务。然后,我们考虑具有更严格的实例保证的认证:对于每个$x^star$,找到一个与$x^star$的最小证书大小一致的证书。与我们的第一个结果形成鲜明对比的是,我们证明这个问题$mathsf{NP}^{mathsf{NP}}$-甚至难以近似。我们得到了一个最优不逼近比,加上在多项式层次的较高层次的少数问题,其中最优不逼近性是已知的。我们的证明涉及到使用位固定分散器进行间隙放大。
{"title":"Certification with an NP Oracle","authors":"Guy Blanc, Caleb M. Koch, Jane Lange, Carmen Strassle, Li-Yang Tan","doi":"10.48550/arXiv.2211.02257","DOIUrl":"https://doi.org/10.48550/arXiv.2211.02257","url":null,"abstract":"In the certification problem, the algorithm is given a function $f$ with certificate complexity $k$ and an input $x^star$, and the goal is to find a certificate of size $le text{poly}(k)$ for $f$'s value at $x^star$. This problem is in $mathsf{NP}^{mathsf{NP}}$, and assuming $mathsf{P} ne mathsf{NP}$, is not in $mathsf{P}$. Prior works, dating back to Valiant in 1984, have therefore sought to design efficient algorithms by imposing assumptions on $f$ such as monotonicity. Our first result is a $mathsf{BPP}^{mathsf{NP}}$ algorithm for the general problem. The key ingredient is a new notion of the balanced influence of variables, a natural variant of influence that corrects for the bias of the function. Balanced influences can be accurately estimated via uniform generation, and classic $mathsf{BPP}^{mathsf{NP}}$ algorithms are known for the latter task. We then consider certification with stricter instance-wise guarantees: for each $x^star$, find a certificate whose size scales with that of the smallest certificate for $x^star$. In sharp contrast with our first result, we show that this problem is $mathsf{NP}^{mathsf{NP}}$-hard even to approximate. We obtain an optimal inapproximability ratio, adding to a small handful of problems in the higher levels of the polynomial hierarchy for which optimal inapproximability is known. Our proof involves the novel use of bit-fixing dispersers for gap amplification.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"205 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122690481","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-02DOI: 10.48550/arXiv.2211.01248
L. N. Coregliano, F. G. Jeronimo, C. Jones
Determining the maximum size $A_2(n,d)$ of a binary code of blocklength $n$ and distance $d$ remains an elusive open question even when restricted to the important class of linear codes. Recently, two linear programming hierarchies extending Delsarte's LP were independently proposed to upper bound $A_2^{text{Lin}}(n,d)$ (the analogue of $A_2(n,d)$ for linear codes). One of these hierarchies, by the authors, was shown to be approximately complete in the sense that the hierarchy converges to $A_2^{text{Lin}}(n,d)$ as the level grows beyond $n^2$. Despite some structural similarities, not even approximate completeness was known for the other hierarchy by Loyfer and Linial. In this work, we prove that both hierarchies recover the exact value of $A_2^{text{Lin}}(n,d)$ at level $n$. We also prove that at this level the polytope of Loyfer and Linial is integral.Even though these hierarchies seem less powerful than general hierarchies such as Sum-of-Squares, we show that they have enough structure to yield exact completeness via pseudoprobabilities.
{"title":"Exact Completeness of LP Hierarchies for Linear Codes","authors":"L. N. Coregliano, F. G. Jeronimo, C. Jones","doi":"10.48550/arXiv.2211.01248","DOIUrl":"https://doi.org/10.48550/arXiv.2211.01248","url":null,"abstract":"Determining the maximum size $A_2(n,d)$ of a binary code of blocklength $n$ and distance $d$ remains an elusive open question even when restricted to the important class of linear codes. Recently, two linear programming hierarchies extending Delsarte's LP were independently proposed to upper bound $A_2^{text{Lin}}(n,d)$ (the analogue of $A_2(n,d)$ for linear codes). One of these hierarchies, by the authors, was shown to be approximately complete in the sense that the hierarchy converges to $A_2^{text{Lin}}(n,d)$ as the level grows beyond $n^2$. Despite some structural similarities, not even approximate completeness was known for the other hierarchy by Loyfer and Linial. In this work, we prove that both hierarchies recover the exact value of $A_2^{text{Lin}}(n,d)$ at level $n$. We also prove that at this level the polytope of Loyfer and Linial is integral.Even though these hierarchies seem less powerful than general hierarchies such as Sum-of-Squares, we show that they have enough structure to yield exact completeness via pseudoprobabilities.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129198478","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-02DOI: 10.48550/arXiv.2211.01443
Pasin Manurangsi
We study the complexity of computing (and approximating) VC Dimension and Littlestone's Dimension when we are given the concept class explicitly. We give a simple reduction from Maximum (Unbalanced) Biclique problem to approximating VC Dimension and Littlestone's Dimension. With this connection, we derive a range of hardness of approximation results and running time lower bounds. For example, under the (randomized) Gap-Exponential Time Hypothesis or the Strongish Planted Clique Hypothesis, we show a tight inapproximability result: both dimensions are hard to approximate to within a factor of $o(log n)$ in polynomial-time. These improve upon constant-factor inapproximability results from [Manurangsi and Rubinstein, COLT 2017].
{"title":"Improved Inapproximability of VC Dimension and Littlestone's Dimension via (Unbalanced) Biclique","authors":"Pasin Manurangsi","doi":"10.48550/arXiv.2211.01443","DOIUrl":"https://doi.org/10.48550/arXiv.2211.01443","url":null,"abstract":"We study the complexity of computing (and approximating) VC Dimension and Littlestone's Dimension when we are given the concept class explicitly. We give a simple reduction from Maximum (Unbalanced) Biclique problem to approximating VC Dimension and Littlestone's Dimension. With this connection, we derive a range of hardness of approximation results and running time lower bounds. For example, under the (randomized) Gap-Exponential Time Hypothesis or the Strongish Planted Clique Hypothesis, we show a tight inapproximability result: both dimensions are hard to approximate to within a factor of $o(log n)$ in polynomial-time. These improve upon constant-factor inapproximability results from [Manurangsi and Rubinstein, COLT 2017].","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115641310","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-16DOI: 10.48550/arXiv.2210.08649
Parikshit Gopalan, Lunjia Hu, Michael P. Kim, Omer Reingold, Udi Wieder
We present a new perspective on loss minimization and the recent notion of Omniprediction through the lens of Outcome Indistingusihability. For a collection of losses and hypothesis class, omniprediction requires that a predictor provide a loss-minimization guarantee simultaneously for every loss in the collection compared to the best (loss-specific) hypothesis in the class. We present a generic template to learn predictors satisfying a guarantee we call Loss Outcome Indistinguishability. For a set of statistical tests--based on a collection of losses and hypothesis class--a predictor is Loss OI if it is indistinguishable (according to the tests) from Nature's true probabilities over outcomes. By design, Loss OI implies omniprediction in a direct and intuitive manner. We simplify Loss OI further, decomposing it into a calibration condition plus multiaccuracy for a class of functions derived from the loss and hypothesis classes. By careful analysis of this class, we give efficient constructions of omnipredictors for interesting classes of loss functions, including non-convex losses. This decomposition highlights the utility of a new multi-group fairness notion that we call calibrated multiaccuracy, which lies in between multiaccuracy and multicalibration. We show that calibrated multiaccuracy implies Loss OI for the important set of convex losses arising from Generalized Linear Models, without requiring full multicalibration. For such losses, we show an equivalence between our computational notion of Loss OI and a geometric notion of indistinguishability, formulated as Pythagorean theorems in the associated Bregman divergence. We give an efficient algorithm for calibrated multiaccuracy with computational complexity comparable to that of multiaccuracy. In all, calibrated multiaccuracy offers an interesting tradeoff point between efficiency and generality in the omniprediction landscape.
{"title":"Loss Minimization through the Lens of Outcome Indistinguishability","authors":"Parikshit Gopalan, Lunjia Hu, Michael P. Kim, Omer Reingold, Udi Wieder","doi":"10.48550/arXiv.2210.08649","DOIUrl":"https://doi.org/10.48550/arXiv.2210.08649","url":null,"abstract":"We present a new perspective on loss minimization and the recent notion of Omniprediction through the lens of Outcome Indistingusihability. For a collection of losses and hypothesis class, omniprediction requires that a predictor provide a loss-minimization guarantee simultaneously for every loss in the collection compared to the best (loss-specific) hypothesis in the class. We present a generic template to learn predictors satisfying a guarantee we call Loss Outcome Indistinguishability. For a set of statistical tests--based on a collection of losses and hypothesis class--a predictor is Loss OI if it is indistinguishable (according to the tests) from Nature's true probabilities over outcomes. By design, Loss OI implies omniprediction in a direct and intuitive manner. We simplify Loss OI further, decomposing it into a calibration condition plus multiaccuracy for a class of functions derived from the loss and hypothesis classes. By careful analysis of this class, we give efficient constructions of omnipredictors for interesting classes of loss functions, including non-convex losses. This decomposition highlights the utility of a new multi-group fairness notion that we call calibrated multiaccuracy, which lies in between multiaccuracy and multicalibration. We show that calibrated multiaccuracy implies Loss OI for the important set of convex losses arising from Generalized Linear Models, without requiring full multicalibration. For such losses, we show an equivalence between our computational notion of Loss OI and a geometric notion of indistinguishability, formulated as Pythagorean theorems in the associated Bregman divergence. We give an efficient algorithm for calibrated multiaccuracy with computational complexity comparable to that of multiaccuracy. In all, calibrated multiaccuracy offers an interesting tradeoff point between efficiency and generality in the omniprediction landscape.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"118 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130170083","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-13DOI: 10.48550/arXiv.2210.06882
Fabien Dufoulon, Y. Emek, R. Gelles
Constructing a shortest path between two network nodes is a fundamental task in distributed computing. This work develops schemes for the construction of shortest paths in randomized beeping networks between a predetermined source node and an arbitrary set of destination nodes. Our first scheme constructs a (single) shortest path to an arbitrary destination in $O (D loglog n + log^3 n)$ rounds with high probability. Our second scheme constructs multiple shortest paths, one per each destination, in $O (D log^2 n + log^3 n)$ rounds with high probability. Our schemes are based on a reduction of the above shortest path construction tasks to a decomposition of hypergraphs into bipartite hypergraphs: We develop a beeping procedure that partitions the (polynomially-large) hyperedge set of a hypergraph $H = (V_H, E_H)$ into $k = Theta (log^2 n)$ disjoint subsets $F_1 cup cdots cup F_k = E_H$ such that the (sub-)hypergraph $(V_H, F_i)$ is bipartite in the sense that there exists a vertex subset $U subseteq V$ such that $|U cap e| = 1$ for every $e in F_i$. This procedure turns out to be instrumental in speeding up shortest path constructions under the beeping model.
构造两个网络节点之间的最短路径是分布式计算的一个基本任务。本工作开发了随机蜂鸣网络中在预定源节点和任意目标节点之间构建最短路径的方案。我们的第一个方案以高概率在$O (D loglog n + log^3 n)$轮中构造一个(单个)到任意目的地的最短路径。我们的第二个方案构建了多个最短路径,每个目的地一条,在$O (D log^2 n + log^3 n)$轮中具有高概率。我们的方案基于将上述最短路径构建任务简化为将超图分解为二部超图:我们开发了一个蜂鸣声过程,将超图$H = (V_H, E_H)$的(多项式大)超边集划分为$k = Theta (log^2 n)$不相交的子集$F_1 cup cdots cup F_k = E_H$,使得(子)超图$(V_H, F_i)$在存在一个顶点子集$U subseteq V$的意义上是二部的,使得$|U cap e| = 1$对于每个$e in F_i$。在蜂鸣声模型下,这一过程有助于加快最短路径的构建。
{"title":"Beeping Shortest Paths via Hypergraph Bipartite Decomposition","authors":"Fabien Dufoulon, Y. Emek, R. Gelles","doi":"10.48550/arXiv.2210.06882","DOIUrl":"https://doi.org/10.48550/arXiv.2210.06882","url":null,"abstract":"Constructing a shortest path between two network nodes is a fundamental task in distributed computing. This work develops schemes for the construction of shortest paths in randomized beeping networks between a predetermined source node and an arbitrary set of destination nodes. Our first scheme constructs a (single) shortest path to an arbitrary destination in $O (D loglog n + log^3 n)$ rounds with high probability. Our second scheme constructs multiple shortest paths, one per each destination, in $O (D log^2 n + log^3 n)$ rounds with high probability. Our schemes are based on a reduction of the above shortest path construction tasks to a decomposition of hypergraphs into bipartite hypergraphs: We develop a beeping procedure that partitions the (polynomially-large) hyperedge set of a hypergraph $H = (V_H, E_H)$ into $k = Theta (log^2 n)$ disjoint subsets $F_1 cup cdots cup F_k = E_H$ such that the (sub-)hypergraph $(V_H, F_i)$ is bipartite in the sense that there exists a vertex subset $U subseteq V$ such that $|U cap e| = 1$ for every $e in F_i$. This procedure turns out to be instrumental in speeding up shortest path constructions under the beeping model.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130247684","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-12DOI: 10.48550/arXiv.2210.05885
Adrian She, H. Yuen
We study unitary property testing, where a quantum algorithm is given query access to a black-box unitary and has to decide whether it satisfies some property. In addition to containing the standard quantum query complexity model (where the unitary encodes a binary string) as a special case, this model contains"inherently quantum"problems that have no classical analogue. Characterizing the query complexity of these problems requires new algorithmic techniques and lower bound methods. Our main contribution is a generalized polynomial method for unitary property testing problems. By leveraging connections with invariant theory, we apply this method to obtain lower bounds on problems such as determining recurrence times of unitaries, approximating the dimension of a marked subspace, and approximating the entanglement entropy of a marked state. We also present a unitary property testing-based approach towards an oracle separation between $mathsf{QMA}$ and $mathsf{QMA(2)}$, a long standing question in quantum complexity theory.
{"title":"Unitary property testing lower bounds by polynomials","authors":"Adrian She, H. Yuen","doi":"10.48550/arXiv.2210.05885","DOIUrl":"https://doi.org/10.48550/arXiv.2210.05885","url":null,"abstract":"We study unitary property testing, where a quantum algorithm is given query access to a black-box unitary and has to decide whether it satisfies some property. In addition to containing the standard quantum query complexity model (where the unitary encodes a binary string) as a special case, this model contains\"inherently quantum\"problems that have no classical analogue. Characterizing the query complexity of these problems requires new algorithmic techniques and lower bound methods. Our main contribution is a generalized polynomial method for unitary property testing problems. By leveraging connections with invariant theory, we apply this method to obtain lower bounds on problems such as determining recurrence times of unitaries, approximating the dimension of a marked subspace, and approximating the entanglement entropy of a marked state. We also present a unitary property testing-based approach towards an oracle separation between $mathsf{QMA}$ and $mathsf{QMA(2)}$, a long standing question in quantum complexity theory.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130265864","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-09-29DOI: 10.4230/LIPIcs.ITCS.2023.64
Sabee Grewal, Vishnu Iyer, William Kretschmer, Daniel Liang
We show that quantum states with"low stabilizer complexity"can be efficiently distinguished from Haar-random. Specifically, given an $n$-qubit pure state $|psirangle$, we give an efficient algorithm that distinguishes whether $|psirangle$ is (i) Haar-random or (ii) a state with stabilizer fidelity at least $frac{1}{k}$ (i.e., has fidelity at least $frac{1}{k}$ with some stabilizer state), promised that one of these is the case. With black-box access to $|psirangle$, our algorithm uses $O!left( k^{12} log(1/delta)right)$ copies of $|psirangle$ and $O!left(n k^{12} log(1/delta)right)$ time to succeed with probability at least $1-delta$, and, with access to a state preparation unitary for $|psirangle$ (and its inverse), $O!left( k^{3} log(1/delta)right)$ queries and $O!left(n k^{3} log(1/delta)right)$ time suffice. As a corollary, we prove that $omega(log(n))$ $T$-gates are necessary for any Clifford+$T$ circuit to prepare computationally pseudorandom quantum states, a first-of-its-kind lower bound.
{"title":"Low-Stabilizer-Complexity Quantum States Are Not Pseudorandom","authors":"Sabee Grewal, Vishnu Iyer, William Kretschmer, Daniel Liang","doi":"10.4230/LIPIcs.ITCS.2023.64","DOIUrl":"https://doi.org/10.4230/LIPIcs.ITCS.2023.64","url":null,"abstract":"We show that quantum states with\"low stabilizer complexity\"can be efficiently distinguished from Haar-random. Specifically, given an $n$-qubit pure state $|psirangle$, we give an efficient algorithm that distinguishes whether $|psirangle$ is (i) Haar-random or (ii) a state with stabilizer fidelity at least $frac{1}{k}$ (i.e., has fidelity at least $frac{1}{k}$ with some stabilizer state), promised that one of these is the case. With black-box access to $|psirangle$, our algorithm uses $O!left( k^{12} log(1/delta)right)$ copies of $|psirangle$ and $O!left(n k^{12} log(1/delta)right)$ time to succeed with probability at least $1-delta$, and, with access to a state preparation unitary for $|psirangle$ (and its inverse), $O!left( k^{3} log(1/delta)right)$ queries and $O!left(n k^{3} log(1/delta)right)$ time suffice. As a corollary, we prove that $omega(log(n))$ $T$-gates are necessary for any Clifford+$T$ circuit to prepare computationally pseudorandom quantum states, a first-of-its-kind lower bound.","PeriodicalId":123734,"journal":{"name":"Information Technology Convergence and Services","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127773874","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}