Searcharxiv⌕ Search

arXiv subjects

Mario Szegedy

Publications and source records attributed to Mario Szegedy.

At least 19 recordsLinked to original sources
arXiv2026

Debate is efficient with your time

AI safety via debate uses two competing models to help a human judge verify complex computational tasks. Previous work has established what problems debate can solve in principle, but has not analysed the practical cost of human oversight: how many queries must the judge make to the debate transcript? We introduce Debate Query Complexity}(DQC), the minimum number of bits a verifier must inspect to correctly decide a debate. Surprisingly, we find that PSPACE/poly (the class of problems which debate can efficiently decide) is precisely the class of functions decidable with O(log n) queries. This characterisation shows that debate is remarkably query-efficient: even for highly complex problems, logarithmic oversight suffices. We also establish that functions depending on all their input bits require Omega(log n) queries, and that any function computable by a circuit of size s satisfies DQC(f) <= log(s) + 3. Interestingly, this last result implies that proving DQC lower bounds of log(n) + 6 for languages in P would yield new circuit lower bounds, connecting debate query complexity to central questions in circuit complexity.

cs.AI↗
arXiv2025

On the Brittleness of LLMs: A Journey around Set Membership

Large language models (LLMs) achieve superhuman performance on complex reasoning tasks, yet often fail on much simpler problems, raising concerns about their reliability and interpretability. We investigate this paradox through a focused study with two key design features: simplicity, to expose basic failure modes, and scale, to enable comprehensive controlled experiments. We focus on set membership queries -- among the most fundamental forms of reasoning -- using tasks like ``Is apple an element of the set \{pear, plum, apple, raspberry\}?''. We conduct a systematic empirical evaluation across prompt phrasing, semantic structure, element ordering, and model choice. Our large-scale analysis reveals that LLM performance on this elementary task is consistently brittle, and unpredictable across all dimensions, suggesting that the models' ``understanding'' of the set concept is fragmented and convoluted at best. Our work demonstrates that the large-scale experiments enabled by the simplicity of the problem allow us to map and analyze the failure modes comprehensively, making this approach a valuable methodology for LLM evaluation in general.

cs.CL↗
arXiv2025

Quantum speedup of non-linear Monte Carlo problems

The mean of a random variable can be understood as a linear functional on the space of probability distributions. Quantum computing is known to provide a quadratic speedup over classical Monte Carlo methods for mean estimation. In this paper, we investigate whether a similar quadratic speedup is achievable for estimating non-linear functionals of probability distributions. We propose a quantum-inside-quantum Monte Carlo algorithm that achieves such a speedup for a broad class of non-linear estimation problems, including nested conditional expectations and stochastic optimization. Our algorithm improves upon the direct application of the quantum multilevel Monte Carlo algorithm introduced by An et al. (2021). The existing lower bound indicates that our algorithm is optimal up polylogarithmic factors. A key innovation of our approach is a new sequence of multilevel Monte Carlo approximations specifically designed for quantum computing, which is central to the algorithm's improved performance.

quant-ph↗
arXiv2025

Nephrobase Cell+: Multimodal Single-Cell Foundation Model for Decoding Kidney Biology

Background: Large foundation models have revolutionized single-cell analysis, yet no kidney-specific model currently exists, and it remains unclear whether organ-focused models can outperform generalized models. The kidney's complex cellular architecture further complicate integration of large-scale omics data, where current frameworks trained on limited datasets struggle to correct batch effects, capture cross-modality variation, and generalize across species. Methods: We developed Nephrobase Cell+, the first kidney-focused large foundation model, pretrained on ~100 billion tokens from ~39.5 million single-cell and single-nucleus profiles across 4,319 samples. Nephrobase Cell+ uses a transformer-based encoder-decoder architecture with gene-token cross-attention and a mixture-of-experts module for scalable representation learning. Results: Nephrobase Cell+ sets a new benchmark for kidney single-cell analysis. It produces tightly clustered, biologically coherent embeddings in human and mouse kidneys, far surpassing previous foundation models such as Geneformer, scGPT, and UCE, as well as traditional methods such as PCA and autoencoders. It achieves the highest cluster concordance and batch-mixing scores, effectively removing donor/assay batch effects while preserving cell-type structure. Cross-species evaluation shows superior alignment of homologous cell types and >90% zero-shot annotation accuracy for major kidney lineages in both human and mouse. Even its 1B-parameter and 500M variants consistently outperform all existing models. Conclusions: Nephrobase Cell+ delivers a unified, high-fidelity representation of kidney biology that is robust, cross-species transferable, and unmatched by current single-cell foundation models, offering a powerful resource for kidney genomics and disease research.

q-bio.GN↗
arXiv2025

Connecting Quantum Computing with Classical Stochastic Simulation

This tutorial paper introduces quantum approaches to Monte Carlo computation with applications in computational finance. We outline the basics of quantum computing using Grover's algorithm for unstructured search to build intuition. We then move slowly to amplitude estimation problems and applications to counting and Monte Carlo integration, again using Grover-type iterations. A hands-on Python/Qiskit implementation illustrates these concepts applied to finance. The paper concludes with a discussion on current challenges in scaling quantum simulation techniques.

quant-ph↗
arXiv2024

Quantum Locally Testable Code with Constant Soundness

In this paper, we present two constructions of quantum locally testable codes (QLTC) with constant soundness. In the first approach, we introduce an operation called check product, and show how this operation gives rise to QLTCs of constant soundness, constant rate, and distance scaling with locality. In the second approach, we consider hypergraph product of a quantum code and a classical repetition code, and observe a special case in which the soundness of component codes is preserved. This insight leads us to construct QLTCs of constant soundness, scalable rate and distance, and constant average locality. Our work marks a step towards constructing QLTCs of high soundness and distance, which would give a different construction to the No Low-Energy Trivial States (NLTS) theorem.

cs.IT↗
arXiv2024

Generating $k$ EPR-pairs from an $n$-party resource state

Motivated by quantum network applications over classical channels, we initiate the study of $n$-party resource states from which LOCC protocols can create EPR-pairs between any $k$ disjoint pairs of parties. We give constructions of such states where $k$ is not too far from the optimal $n/2$ while the individual parties need to hold only a constant number of qubits. In the special case when each party holds only one qubit, we describe a family of $n$-qubit states with $k$ proportional to $\log n$ based on Reed-Muller codes, as well as small numerically found examples for $k=2$ and $k=3$. We also prove some lower bounds, for example showing that if $k=n/2$ then the parties must have at least $Ω(\log\log n)$ qubits each.

quant-ph↗
arXiv2024

Quadratic Speed-up in Infinite Variance Quantum Monte Carlo

In this study, we give an extension of Montanaro's arXiv/archive:1504.06987 quantum Monte Carlo method, tailored for computing expected values of random variables that exhibit infinite variance. This addresses a challenge in analyzing heavy-tailed distributions, which are commonly encountered in various scientific and engineering fields. Our quantum algorithm efficiently estimates means for variables with a finite $(1+δ)^{\text{th}}$ moment, where $δ$ lies between 0 and 1. It provides a quadratic speedup over the classical Monte Carlo method in both the accuracy parameter $ε$ and the specified moment of the distribution. We establish both classical and quantum lower bounds, showcasing the near-optimal efficiency of our algorithm among quantum methods. Our work focuses not on creating new algorithms, but on analyzing the execution of existing algorithms with available additional information about the random variable. Additionally, we categorize these scenarios and demonstrate a hierarchy in the types of supplementary information that can be provided.

quant-ph↗
arXiv2023

Rubik Tables and Object Rearrangement

A great number of robotics applications demand the rearrangement of many mobile objects, e.g., organizing products on shelves, shuffling containers at shipping ports, reconfiguring fleets of mobile robots, and so on. To boost the throughput in systems designed for solving these rearrangement problems, it is essential to minimize the number of atomic operations, e.g., the pick-n-places of individual objects. However, this optimization task poses a rather difficult challenge due to complex inter-dependency between objects, especially in high-density settings. In tackling the aforementioned challenges, we develop a novel algorithmic tool, Rubik Tables, that provides a clean abstraction of object rearrangement problems as the proxy problem of shuffling items stored in a table or lattice. In its basic form, a Rubik Table is an $n\times n$ table containing $n^2$ items. We show that the reconfiguration of items in such a Rubik Table can be achieved using at most $2n$ column/row shuffles in the partially labeled setting, where each column (resp., row) shuffle may arbitrarily permute the items stored in a column (resp., row) of the table. When items are fully distinguishable, additional $n$ shuffles are needed. Rubik Tables allow many generalizations, e.g., to higher dimensions. Using Rubik Table, we have designed a first constant-factor optimal algorithm for stack rearrangement problems. We show that, for $nd$ items stored in $n$ stacks of depth $d$ each, using one empty stack as the swap space, $O(nd)$ stack pop-push operations are sufficient for an arbitrary reconfiguration of the stacks where $d \le n^{\frac{m}{2}}$ for arbitrary fixed $m >0$. Rubik Table results also allow the development of constant-factor optimal solutions for solving multi-robot motion planning problems under extreme robot density. These algorithms based on Rubik Table results run in low-polynomial time.

cs.RO↗
arXiv2022

Repeated Averages on Graphs

Sourav Chatterjee, Persi Diaconis, Allan Sly and Lingfu Zhang, prompted by a question of Ramis Movassagh, renewed the study of a process proposed in the early 1980s by Jean Bourgain. A state vector $v \in \mathbb R^n$, labeled with the vertices of a connected graph, $G$, changes in discrete time steps following the simple rule that at each step a random edge $(i,j)$ is picked and $v_i$ and $v_j$ are both replaced by their average $(v_i+v_j)/2$. It is easy to see that the value associated with each vertex converges to $1/n$. The question was how quickly will $v$ be $ε$-close to uniform in the $L^{1}$ norm in the case of the complete graph, $K_{n}$, when $v$ is initialized as a standard basis vector that takes the value 1 on one coordinate, and zeros everywhere else. They have established a sharp cutoff of $\frac{1}{2\log 2}n\log n + O(n\sqrt{\log n})$. Our main result is to prove, that $\frac{(1-ε)}{2\log2}n\log n-O(n)$ is a general lower bound for all connected graphs on $n$ nodes. We also get sharp magnitude of $t_{ε,1}$ for several important families of graphs, including star, expander, dumbbell, and cycle. In order to establish our results we make several observations about the process, such as the worst case initialization is always a standard basis vector. Our results add to the body of work of Aldous, Aldous and Lanoue, Quattropani and Sau, Cao, Olshevsky and Tsitsiklis, and others. The renewed interest is due to an analogy to a question related to the Google's supremacy circuit. For the proof of our main theorem we employ a concept that we call 'augmented entropy function' which may find independent interest in the computer science and probability theory communities.

math.PR↗
arXiv2022

Budgeted Steiner Networks: Three Terminals with Equal Path Weights

Given a set of terminals in 2D/3D, the network with the shortest total length that connects all terminals is a Steiner tree. On the other hand, with enough budget, every terminal can be connected to every other terminals via a straight edge, yielding a complete graph over all terminals. In this work, we study a generalization of Steiner trees asking what happens in between these two extremes. Focusing on three terminals with equal pairwise path weights, we characterize the full evolutionary pathway between the Steiner tree and the complete graph, which contains intriguing intermediate structures.

cs.CG↗
arXiv2020

Classical Simulation of Quantum Supremacy Circuits

It is believed that random quantum circuits are difficult to simulate classically. These have been used to demonstrate quantum supremacy: the execution of a computational task on a quantum computer that is infeasible for any classical computer. The task underlying the assertion of quantum supremacy by Arute et al. (Nature, 574, 505--510 (2019)) was initially estimated to require Summit, the world's most powerful supercomputer today, approximately 10,000 years. The same task was performed on the Sycamore quantum processor in only 200 seconds. In this work, we present a tensor network-based classical simulation algorithm. Using a Summit-comparable cluster, we estimate that our simulator can perform this task in less than 20 days. On moderately-sized instances, we reduce the runtime from years to minutes, running several times faster than Sycamore itself. These estimates are based on explicit simulations of parallel subtasks, and leave no room for hidden costs. The simulator's key ingredient is identifying and optimizing the "stem" of the computation: a sequence of pairwise tensor contractions that dominates the computational cost. This orders-of-magnitude reduction in classical simulation time, together with proposals for further significant improvements, indicates that achieving quantum supremacy may require a period of continuing quantum hardware developments without an unequivocal first demonstration.

quant-ph↗
arXiv2020

Finding Angles for Quantum Signal Processing with Machine Precision

We describe an algorithm for finding angle sequences in quantum signal processing, with a novel component we call halving based on a new algebraic uniqueness theorem, and another we call capitalization. We present both theoretical and experimental results that demonstrate the performance of the new algorithm. In particular, these two algorithmic ideas allow us to find sequences of more than 3000 angles within 5 minutes for important applications such as Hamiltonian simulation, all in standard double precision arithmetic. This is native to almost all hardware.

quant-ph↗
arXiv2019

What do QAOA energies reveal about graphs?

Quantum Approximate Optimization Algorithm (QAOA) is a hybrid classical-quantum algorithm to approximately solve NP optimization problems such as MAX-CUT. We describe a new application area of QAOA circuits: graph structure discovery. We omit the time-consuming parameter-optimization phase and utilize the dependence of QAOA energy on the graph structure for randomly or judiciously chosen parameters to learn about graphs. In the first part, Following up on Wang et. al. and Brandao et. al. we give explicit formulas. We show that the layer-one QAOA energy for the MAX-CUT problem for three regular graphs carries exactly the information: {\em (# of vertices, # of triangles)}. We have calculated our explicit formulas differently from \cite{wang2018quantum}, by developing the notion of the $U$ polynomial of a graph $G$. Many of our discoveries can be interpreted as computing $U(G)$ under various restrictions. The most basic question when comparing the structure of two graphs is if they are isomorphic or not. We find that the QAOA energies separate all non-isomorphic three-regular graphs up to size 18, all strongly regular graphs up to size 26 and the Praust and the smallest Miyazaki examples. We observe that the QAOA energy values can be also used as a proxy to how much graphs differ. Unfortunately, we have also found a sequence of non-isomorphic pairs of graphs, for which the energy gap seems to shrink at an exponential rate as the size grows. Our negative findings however come with a surprise: if the QAOA energies do not measurably separate between two graphs, then both of their energy landscapes must be extremely flat (indistinguishable from constant), already when the number of QAOA layers is intermediately large. This holds due to a remarkable uncoupling phenomenon that we have only deduced from computer simulation.

quant-ph↗
arXiv2019

Alibaba Cloud Quantum Development Platform: Large-Scale Classical Simulation of Quantum Circuits

We report, in a sequence of notes, our work on the Alibaba Cloud Quantum Development Platform(AC-QDP). AC-QDP provides a set of tools for aiding the development of both quantum computing algorithms and quantum processors, and is powered by a large-scale classical simulator deployed on Alibaba Cloud. In this note, we report the computational experiments demonstrating the classical simulation capability of AC-QDP. We use as a benchmark the random quantum circuits designed for Google's Bristlecone QPU {\cite{GRCS}}. We simulate Bristlecone-70 circuits with depth $1 + 32 + 1$ in $0.43$ second per amplitude, using $1449$ Alibaba Cloud Elastic Computing Service (ECS) instances, each with $88$ Intel Xeon(Skylake) Platinum 8163 vCPU cores @ 2.5 GHz and $160$ gigabytes of memory. By comparison, the previously best reported results for the same tasks are $104$ and $135$ seconds, using NASA's HPC Pleiades and Electra systems, respectively ({arXiv:1811.09599}). Furthermore, we report simulations of Bristlecone-70 with depth $1+36+1$ and depth $1+40+1$ in $5.6$ and $580.7$ seconds per amplitude, respectively. To the best of our knowledge, these are the first successful simulations of instances at these depths.

quant-ph↗
arXiv2019

Alibaba Cloud Quantum Development Platform: Applications to Quantum Algorithm Design

We report our work on the Alibaba Cloud Quantum Development Platform (AC-QDP). The capability of AC-QDP's computational engine was already reported in \cite{CZH+18, ZHN+19}.In this follow-up article, we demonstrate with figures how AC-QDP helps in testing large-scale quantum algorithms (currently within the QAOA framework). We give new benchmark results on regular graphs. AC-QDP's QAOA framework can simulate thousands of qubits for up to $4$ layers. Then we discuss two interesting use cases we have implemented on the platform: 1. Optimal QAOA sequences for small-cycle free graphs; 2. Graph structure discovery.

quant-ph↗
arXiv2019

The Power Light Cone of the Discrete Bak-Sneppen, Contact and other local processes

We consider a class of random processes on graphs that include the discrete Bak-Sneppen (DBS) process and the several versions of the contact process (CP), with a focus on the former. These processes are parametrized by a probability $0\leq p \leq 1$ that controls a local update rule. Numerical simulations reveal a phase transition when $p$ goes from 0 to 1. Analytically little is known about the phase transition threshold, even for one-dimensional chains. In this article we consider a power-series approach based on representing certain quantities, such as the survival probability or the expected number of steps per site to reach the steady state, as a power-series in $p$. We prove that the coefficients of those power series stabilize as the length $n$ of the chain grows. This is a phenomenon that has been used in the physics community but was not yet proven. We show that for local events $A,B$ of which the support is a distance $d$ apart we have $\mathrm{cor}(A,B) = \mathcal{O}(p^d)$. The stabilization allows for the (exact) computation of coefficients for arbitrary large systems which can then be analyzed using the wide range of existing methods of power series analysis.

math.PR↗