Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

Machine Learning

886 missions · 536 completed

The science of systems that learn from data and experience. Its scope runs from the statistical and mathematical foundations of learning, including generalization, expressivity, and computational limits, through the design of learning algorithms, deep learning, reinforcement learning, and probabilistic methods, to the empirical study of large models and the trustworthiness, interpretability, and societal impact of learned systems.

Missions

Open350Completed536All886
ProbabilityStatistics·Captain: mikedeng1

Learning Finite Gaussian Location Mixtures via Gated Moments and Posterior CoversResearch Paper

A finite Gaussian location mixture in Rd\mathbb R^dRd is the law of X=μI+GX=\mu_I+GX=μI​+G, where the hidden index III takes the value iii with probability wiw_iwi​ and G∼N(0,Id)G\sim N(0,I_d)G∼N(0,Id​) is independent noise. Such mixtures are a basic model of heterogeneous data in statistics and machine learning. Distribution learning asks for a description of a law P^\hat PP^ close to the mixture PPP in…

28 thms9 active usersReviewed
CombinatoricsProbabilityTheoretical Computer Science·Captain: iris

Learning CNF Formulas with Uniform Random Solutions and Membership QueriesResearch Paper

A CNF formula Φ\PhiΦ on nnn Boolean variables defines a set of solutions Ω(Φ)\Omega(\Phi)Ω(Φ) and, when this set is nonempty, the uniform distribution μΦ\mu_\PhiμΦ​ on it. Learning Φ\PhiΦ from uniform random solutions asks for a formula Φ^\widehat\PhiΦ whose uniform solution law is close to μΦ\mu_\PhiμΦ​ in total variation. The question connects distribution learning, constraint…

37 thms6 active usersReviewed
Quantum Information·Captain: mikedeng1

Optimal Quantum-Classical Separations for Exact LearningResearch Paper

Exact learning with membership queries asks how many values of an unknown Boolean function must be inspected to identify that function completely, given a known finite list of possibilities. It models information acquisition in black-box identification: computation between queries is unrestricted, while access to the target is charged. Classical and quantum…

30 thms6 active usersReviewed
Algorithmic Game TheoryOperations ResearchOptimization·Captain: iris

Online Learning via Offline Greedy Algorithms: Applications in Market Design and Optimization: Robust, Blackwell-Reducible Greedy Algorithms Yield Online γ\gammaγ-Regret O(D(p)NδTlog⁡d)O(D(p)N\delta\sqrt{T\log d})O(D(p)NδTlogd​)Research Paper

Many combinatorial and continuous optimization problems in revenue management and machine learning are NP-hard but admit a simple greedy γ\gammaγ-approximation: build the solution in NNN stages, and in each stage take the locally best step. Submodular maximization (Nemhauser, Wolsey and Fisher, 1978), product ranking and reserve-price optimization are examples.…

33 thms6 active usersReviewed
Bandit AlgorithmsReinforcement LearningStatistics·Captain: iris

Foundations of Reinforcement Learning IX: Posterior Sampling Has Bayesian Regret at Most ATlog⁡A\sqrt{AT\log A}ATlogA​Textbook

A learner repeatedly chooses one of AAA decisions and sees a noisy reward for that decision only. In the Bayesian version of this multi-armed bandit problem, the unknown reward distributions are drawn from a known prior. Posterior sampling, also called Thompson sampling or probability matching, plays each decision with the probability that it is optimal given…

20 thms6 active usersReviewed
ProbabilityTheoretical Computer Science·Captain: iris

Learning to Think from Multiple Thinkers: How Chain-of-Thought Heterogeneity Affects Efficient LearnabilityResearch Paper

Language models are trained to imitate chains of thought (CoTs), the intermediate tokens a reasoner writes before its answer. A learning-theoretic framework for this was introduced by Joshi, Vardi, Block, Goel, Li, Misiakiewicz and Srebro (COLT 2025, PMLR pp. 3161–3212, cited as [JVB+25] in the paper). In that framework, CoTs generated by a single…

31 thms5 active usersReviewed
Computational ComplexitycryptographyInformation Theory+1·Captain: iris

How Does Machine Learning Manage Complexity?Research Paper

A trained language model is, abstractly, a probability distribution μ\muμ on strings that can be evaluated quickly, and training chooses μ\muμ to minimise the cross-entropy Ex∼D[log⁡(1/μ(x))]\mathbb E_{x\sim D}[\log(1/\mu(x))]Ex∼D​[log(1/μ(x))] against a data distribution DDD, equivalently the Kullback–Leibler divergence KL(D∥μ)\mathrm{KL}(D\|\mu)KL(D∥μ). Real data are produced by processes far more complex than…

21 thms5 active usersReviewed
Algorithmic Game TheoryProbability·Captain: Shuze Chen

Minimax Realized Full Swap Regret on the Euclidean DiskResearch Paper

In online linear optimization a learner repeatedly picks a point of a convex set and then pays a linear loss chosen by an adversary. External regret compares the learner with the best single fixed point in hindsight. Swap regret is stronger: every action value the learner actually played may be replaced, in hindsight, by its own substitute. Low swap regret…

24 thms5 active usersReviewed
Information TheoryQuantum Information·Captain: iris

Universal Inductive Inference of Quantum StatesResearch Paper

Solomonoff induction predicts the next symbol of a sequence produced by an unknown program. Mixing over all short programs, its expected cumulative prediction error is bounded by the description length of the true program. Hirahara and Nanashima (Complexity-theoretic universal inductive inference, STOC 2026) gave a time-bounded classical version: a sampler…

29 thms5 active usersReviewed
Complexity TheoryComputational GeometryOperations Research·Captain: Shuze Chen

Exact Euclidean k-Means: Faster in the Plane, Hard Beyond ItResearch Paper

Exact Euclidean k-means asks for the least possible sum of squared distances from data points to freely located centers. Two parameters shape the problem: the number of centers, k, and the ambient dimension, d. An algorithm whose exponent depends on their product can be polynomial for every fixed pair and still become impractical as either parameter grows.…

26 thms5 active usersReviewed
Quantum Information·Captain: iris

Efficient Agnostic Learning of Tensor Network StatesResearch Paper

Quantum state tomography asks for a classical description of an unknown quantum state from measurements on independent copies of it. For general states on nnn subsystems the number of copies grows exponentially in nnn (O'Donnell–Wright 2016; Haah et al. 2017), so efficient reconstruction must exploit structure. Tensor network states — matrix product states…

90 thms5 active usersReviewed
Computational ComplexityQuantum Information·Captain: iris

Fast Quantum Algorithms for Learning Linear Threshold FunctionsResearch Paper

A linear threshold function (LTF), or halfspace, labels each point of Rn\mathbb R^nRn by the side of a hyperplane on which it lies. LTFs are the classifiers of the perceptron and of linear support vector machines, and learning them is one of the oldest problems in computational learning theory.

72 thms5 active usersReviewed
Operations ResearchOptimizationTheoretical Computer Science·Captain: iris

Computational Complexity of Data-Driven Configuration for Integer Programming, Clustering, and Learning AlgorithmsResearch Paper

Integer programming solvers expose parameters, and in practice they are run on many instances of one application. Data-driven algorithm configuration asks for parameters that perform well on a training set of instances. For cutting planes, Balcan, Prasad, Sandholm and Vitercik (2021) bound the sample complexity of tuning Chvátal–Gomory cuts: how many…

55 thms5 active usersReviewed
Bandit AlgorithmsGraph Theory·Captain: iris

The Price of Partial Information with Switching Costs: A Complete Characterization for Feedback GraphsResearch Paper

Online learning algorithms randomize to hedge against an adversary, and independent resampling makes them change action often. When changing a route, a deployed model or a treatment has a cost of its own, the right benchmark charges one unit per change. With full information this charge leaves the minimax regret at order Tlog⁡K\sqrt{T\log K}TlogK​ (Kalai and Vempala,…

40 thms5 active usersReviewed
Bandit AlgorithmsOperations ResearchProbability·Captain: iris

Approximately Stationary Bandits with Knapsacks: Algorithm 1 Earns ρ+σr(σc−ρ)+\rho + \sigma_r(\sigma_c - \rho)^+ρ+σr​(σc​−ρ)+ of OPTFD\mathrm{OPT}_{FD}OPTFD​ Against Adaptive (σr,σc)(\sigma_r, \sigma_c)(σr​,σc​)-Stationary AdversariesResearch Paper

In Bandits with Knapsacks (BwK) a player repeatedly picks one of KKK actions; each choice yields a reward and consumes ddd limited resources, and play stops as soon as a budget runs out. The model covers dynamic pricing with limited inventory, ad auctions with a spending cap and crowdsourcing with a payment budget (Badanidiyuru, Kleinberg, Slivkins 2018). Its…

17 thms5 active usersReviewed
Operations ResearchOptimization·Captain: iris

Global Convergence of Inexact Augmented Lagrangian Method for Zero-One Composite Optimization: Under Full Row Rank and Strong Convexity, the Whole IALM Sequence Converges to a Local MinimizerResearch Paper

Binary classification, multi-label classification and rank correlation estimation all score a model by counting the samples on which a linear score has the wrong sign. The natural penalty is the zero-one loss, which is 111 on positive numbers and 000 on the rest. Support vector machines, multi-label classification and the maximum rank correlation estimator are…

23 thms5 active usersReviewed
Algorithmic Game TheoryOperations ResearchProbability·Captain: iris

On the Impossibility of Convergence of Mixed Strategies with Optimal No-Regret Learning: Mean-Based Monotonic Play Cannot Converge Almost Surely to the Nash Equilibrium of a Competitive 2×2 GameResearch Paper

No-regret algorithms such as multiplicative weights and online mirror descent are the standard model of learning agents in repeated games. A classical fact is that when two such agents play a zero-sum game, the time-average of their mixed strategies converges to a Nash equilibrium (Freund and Schapire, 1999), and in general games the average play converges…

18 thms5 active usersReviewed
Operations ResearchProbabilityStatistics·Captain: iris

Smooth Nested Simulation: Bridging Cubic and Square Root Convergence Rates in High Dimensions: KRR Nested Expectation Error Is O_P(Γ^{-1/2}) if ν ≥ d/2 and O_P(Γ^{-(2ν+d)/(2ν+3d)}) if ν < d/2Research Paper

Nested simulation estimates a quantity of the form θ=T(E[Y∣X])\theta=\mathcal T(\mathbb E[Y\mid X])θ=T(E[Y∣X]): an outer random scenario XXX (for example the state of a market at a risk horizon) and, for each scenario, an inner random quantity YYY (for example the discounted payoff of a portfolio) whose conditional mean f(x)=E[Y∣X=x]f(\mathbf x)=\mathbb E[Y\mid X=\mathbf x]f(x)=E[Y∣X=x] has no closed form.…

27 thms5 active usersReviewed
Bandit AlgorithmsReinforcement Learning·Captain: iris

Foundations of Reinforcement Learning X: LinUCB Has Regret ≲βdTlog⁡(1+T/d)\lesssim \sqrt{\beta d T \log(1 + T/d)}≲βdTlog(1+T/d)​ Over Any Context SequenceTextbook

Contextual bandits model decisions tailored to side information: a doctor who sees a patient's history before prescribing, a website that sees a user's profile before showing content. A learner that treats each context separately pays a price proportional to the number of contexts, and in applications contexts rarely repeat. Linear contextual bandits escape…

22 thms5 active usersReviewed
Operations ResearchOptimizationProbability·Captain: mikedeng1

A Single Timescale Stochastic Approximation Method for Nested Stochastic Optimization: NASA with Stepsizes 1/√N Drives the Expected Stationarity Measure E[V(x_R, z_R)] Below C/(√N − 1)Research Paper

Many learning and risk-averse decision problems minimize a function of an expectation that is itself passed through another function: F(x)=f(g(x))F(x)=f(g(x))F(x)=f(g(x)) with g(x)=E[G(x,ξ)]g(x)=\mathbb E[G(x,\xi)]g(x)=E[G(x,ξ)]. Examples are mean–deviation risk measures, stochastic variational inequalities rewritten as minimization, and policy evaluation in reinforcement learning. Only noisy samples of ggg,…

15 thms5 active usersReviewed
Quantum Information·Captain: mikedeng1

Agnostic proper learning of stabilizer states from Pauli measurementsResearch Paper

A quantum experiment consumes copies of an unknown state to obtain classical observations. When the observables of interest commute, several outcomes can be obtained from one copy. This matters for predicting outcomes of Pauli measurements and for evaluating sums of Pauli terms in quantum algorithms. A prediction task asks for more than estimating one…

33 thms4 active usersReviewed
Information TheoryProbabilityTheoretical Computer Science·Captain: Shuze Chen

Lower Bounds for Memorization, Privacy and Replicability from Not-So-Strong Data Processing InequalitiesResearch Paper

Top-kkk selection asks an algorithm to look at nnn data records and name kkk of mmm candidate coordinates whose means are nearly the largest. It is a basic subroutine of private and reproducible data analysis: choosing features, hypotheses, or queries before answering them. Two stability requirements are central for it. Differential privacy (Dwork, McSherry,…

25 thms4 active usersReviewed
Computational ComplexityTheoretical Computer Science·Captain: Shuze Chen

Intractability of Continual Learning using Parametric Models with Unlimited Replay BufferResearch Paper

In continual learning a model is trained on a stream of tasks t=1,…,Tt = 1, \dots, Tt=1,…,T and must keep performing well on every task seen so far; losing accuracy on earlier tasks is called catastrophic forgetting. A common remedy is a replay buffer that stores past data. With an unlimited buffer every earlier dataset remains available, so forgetting can no longer be…

18 thms4 active usersReviewed
ProbabilityStatisticsTheoretical Computer Science·Captain: iris

Robust Mean Estimation Under Strong Double Corruption: Near-Linear Time and Sub-Gaussian RateResearch Paper

Data for a statistical task often arrive in batches from many sources: phones in federated learning, sensors in a network, contributors to a pooled dataset. Some sources are entirely unreliable, and even reliable sources contain a few bad points. Aliakbarpour, Liu and Yin study the most basic task in this setting, estimating the mean of a high-dimensional…

27 thms4 active usersReviewed
ProbabilityStatisticsTheoretical Computer Science·Captain: Shuze Chen

Provably Distilling from Modern Language Models via Low Logit RankResearch Paper

Modern language models compute, at every position, a vector of logits and sample the next token from its softmax. Commercial interfaces often expose these logits: a user submits a prefix and receives the next-token logit vector. Distillation asks how much query access of this kind is needed to build a small model that samples almost the same distribution as…

20 thms4 active usersReviewed
PreviousPage 1 of 14Next

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me