Practical Quantum Topological Data Analysis with Applications to High-Dimensional Feature Extraction and Time Series Analysis

AI summary

Overview

Research area: Quantum algorithms for topological data analysis (TDA), applied to high-dimensional feature extraction from time-series data.

Technical level: Intermediate to Advanced. The paper assumes familiarity with simplicial complexes, homology, Betti numbers, and the Dirac/Laplacian operator construction used in quantum TDA; the application sections (fMRI and finance) are more accessible.

Scope (one sentence): The paper argues that quantum TDA is better framed as a low-order Laplacian moment extraction problem for downstream machine learning rather than exact Betti number estimation, and supports this with two time-series applications, a moment-based algorithm, resource estimates, and trapped-ion hardware demonstrations.

Note: the manuscript content supplied for this summary is truncated mid-way through Section 3.1.2. Sections covering the financial time-series application in detail and the full quantum algorithm, resource estimates, crossover projections, and hardware results in Section 4 were not available, so those parts are described only as characterized in the abstract and introduction.

What This Paper Is About

Topological data analysis extracts shape-based features (connected components, loops, voids, and their higher-dimensional analogues) from complex data, but classical computation of high-dimensional Betti numbers becomes intractable because the number of possible d-simplices on N vertices grows combinatorially as C(N, d+1). Quantum algorithms promise a way around this bottleneck, yet prior work framed the goal as exact or high-precision Betti number estimation, which made the regime for practical advantage look narrow. This paper reframes quantum TDA as a feature-extraction tool that measures low-order moments of the combinatorial Laplacian as a proxy for high-dimensional topology, and tests whether those features actually help on real prediction tasks.

Key Contributions

1. Demonstration that higher-order TDA features improve predictive performance in two time-series domains. The authors extend prior classical TDA studies, which were restricted to H₀ and H₁ homologies, up to H₄ on established datasets: fMRI-based neurodegenerative disease classification and financial time-series analysis for detecting early indicators of market instability.

2. A moment-based quantum algorithm targeting low-order spectral information of the combinatorial Laplacian. Rather than estimating Betti numbers directly, the method extracts Laplacian-derived observables, including the relative trace, and the authors show these low-order moments are strongly correlated with high-dimensional Betti information even when the relative Betti number is small.

3. Resource estimates and quantum-classical crossover projections, together with explicit circuit constructions, for the moment-based method.

4. Experimental demonstration on Barium-based trapped-ion hardware, described as similar to the forthcoming IonQ Tempo line, extracting Laplacian-derived observables from graph instances with up to 16 nodes and quantitatively comparing them against exact Betti information.

5. A classical baseline study establishing how time-to-solution and memory scale for state-of-the-art TDA software, providing the comparison point for the quantum approach.

Main Findings

• Higher homology features improve classification of Alzheimer's disease from fMRI. On an OASIS subset of 32 AD and 79 healthy individuals (111 patients total), TDA-only features gave 69% accuracy with H₀ alone, rising monotonically as higher homology dimensions were added: 70.5% for H₀:H₁, 72.5% for H₀:H₂, 73.5% for H₀:H₃, and 74% for H₀:H₄.

• The comparison to prior work is explicitly framed as non-comparable. The paper's own Table 1 places these numbers beside Ref. [3] (ADNI, 141 AD and 177 healthy patients; region-wise accuracies of 74.6% for DMN with H₀, 75.4% for FP with H₁, 81% for OP with H₀, 69.9% for SM with H₀, 77.3% for CO with H₁, and 71.6% for CB with H₂) and Ref. [34] (OASIS: 101 AD, 102 healthy, 86.16%; ADNI: 55 AD, 95 healthy, 93.9% with SMOTE augmentation). The authors state these accuracies are not directly comparable because datasets, pre-processing pipelines, atlases, and model classes differ.

• Controlled feature-substitution experiments isolate the effect of higher homology. The neural network architecture feeds all homology features through a shared first layer, and unused homology features are replaced with random features so that the number of trainable parameters is held constant across conditions. This tests whether the higher-dimensional features carry signal beyond simply increasing input dimension.

• Classical TDA time-to-solution follows a clean scaling law. Using giotto-ph on Erdős–Rényi random graphs, with up to 150 CPU nodes of 500 GB each and Betti numbers computed up to d = 10, the time-to-solution collapses onto TTS = a·C(N,d)/d + b with a = 7.6 × 10⁻⁹ and b = −0.037.

• Memory is the tighter classical constraint. Peak memory scales as MaxRSS = a·C(N,d) + b with a = 3.35 × 10⁻⁸ and b = −12.284. This is slightly worse scaling than time-to-solution, so memory may present an earlier barrier than total runtime.

• Classical high-order Betti computation fails on modest graphs. Already for graphs of approximately 80 nodes it becomes difficult to calculate Betti numbers above d = 8. Time-to-solution within giotto-ph does not depend strongly on graph edge density.

• The quantum resource requirement does not appear to demand a large relative Betti number. The authors report that, contrary to conventional wisdom, low-order moments including the relative trace correlate strongly with high-dimensional Betti information even when the relative Betti number is small, which substantially expands the class of simplicial complexes where quantum TDA may be useful.

• Hardware extraction of Laplacian observables was demonstrated on up to 16-node graphs on a Barium development system, with quantitative comparison to exact Betti information. Full experimental numbers were not available in the supplied excerpt.

Methodology in Plain English

The authors build simplicial complexes — generalizations of graphs that include triangles, tetrahedra, and higher-dimensional cliques — from data using Vietoris–Rips filtrations, where a clique is included if all pairwise distances between its points fall below a threshold ε. They then study homology, the algebraic summary of holes in each dimension, and specifically the Betti number β_d, which counts independent d-cycles that are not boundaries of filled-in objects. Computationally, this is expressed through a boundary operator built from fermionic creation and annihilation operators, whose square yields the combinatorial Laplacian; the dimension of that Laplacian's kernel equals the Betti number.

On the application side, they take fMRI scans from the OASIS dataset, reduce each brain to 48 cortical regions using the Harvard–Oxford atlas, and compute a 48 × 48 correlation matrix per scan. Each region's time series is converted, via Takens embedding from the giotto-tda library, into a 10-dimensional point cloud of roughly 150 points. Persistent homology is computed on each point cloud for homology dimensions 0 through 4, and the distance between every pair of regions' persistence diagrams forms a 48 × 48 distance matrix per homology dimension, which then feeds a classifier. These features were compressed to 12 × 12 arrays to limit overfitting in small networks, and evaluated with a two-layer linear neural network (and separately an SVM) using an 80%–20% train–test split, with balanced accuracy averaged over 200 independent runs because the dataset is imbalanced.

On the algorithmic side, they replace the goal of exact Betti number estimation with estimation of low-order moments of the combinatorial Laplacian, including the relative trace, and check empirically how strongly those moments track the exact high-dimensional Betti information. Classical baselines were generated with giotto-ph, which builds on Ripser and GUDHI and uses an edge-collapser pre-processing step to shrink the complex.

Why This Matters

Impact on research: The paper reframes the quantum TDA target from precise Betti number estimation to a feature-extraction primitive. If low-order Laplacian moments carry the relevant topological signal, the algorithm is not restricted to the narrow regime of large relative Betti numbers that earlier complexity analyses implied, which changes where the community should look for practical advantage. The paper is also an explicit example of "bottom-up" quantum application development: identify an algorithmic regime where quantum methods may win, then find applications where the extracted information has real value.

Real-world applications:

• Neurodegenerative disease diagnosis: distinguishing Alzheimer's disease from healthy cognition using fMRI-derived connectivity features, with evidence that homology dimensions beyond H₀ and H₁ add signal.
• Financial market instability detection: using higher-order TDA features on financial time series to flag early indicators of instability.
• General high-dimensional feature extraction: producing compact topological descriptors for downstream AI models on network and point-cloud data where pairwise summaries miss higher-order structure.
• Complex systems modeling: characterizing systems whose large-scale behavior depends on many interacting components, where higher-order interactions beyond pairwise ones matter.

Industry relevance: The work comes from IonQ and ties the algorithm to a specific hardware roadmap, comparing the development Barium system to the forthcoming IonQ Tempo line. Combined with circuit constructions, resource estimates, and crossover projections, the paper is aimed at establishing a practical pathway to quantum utility rather than a purely asymptotic speedup claim.

Future Directions

• Establishing an apples-to-apples classical baseline. The authors explicitly state that a true comparison would require training conventional correlation-based or deep-learning models on the same reduced 111-patient OASIS subset under the same validation protocol, which they did not do.

• Scaling the hardware demonstrations beyond 16-node graphs and demonstrating the crossover point where the quantum moment extraction outperforms classical Betti computation in practice.

• Determining the full boundary of the "small relative Betti number" regime. The claim that large relative Betti numbers are not required opens a broader class of complexes, but the scope of that class is not fully characterized in the available text.

• Extending the moment-based approach to more homology dimensions and denser complexes, given that classical memory scaling as C(N,d) becomes prohibitive before runtime does.

Target Audience

Quantum computing researchers working on near-term applications and quantum machine learning, particularly those interested in combinatorial Laplacian and dequantization-adjacent algorithm design. Also relevant to applied topologists and data scientists who use TDA on time-series and network data, and to industry strategists evaluating where trapped-ion hardware might reach practical utility. Readers need comfort with algebraic topology terminology to follow Section 2, though the application sections are readable with less background.

Authors’ abstract

Topological data analysis (TDA) provides a powerful framework for extracting information about the shape of complex, unstructured data, but the classical cost of computing high dimensional topological features limits its application. Quantum algorithms for TDA offer a route around this bottleneck, yet existing approaches typically focus on exact or high precision Betti number estimation, making the regime for practical quantum advantage appear narrow. Here, we instead frame quantum TDA as a feature-extraction method for downstream data analysis by extracting low-order spectral information from the combinatorial Laplacian as a proxy for high-dimensional topology. We support this perspective from both the application and algorithmic sides. First, we show that higher-order TDA features improve predictive performance in two time-series applications: functional MRI analysis for neurodegenerative disease classification and financial time-series analysis for identifying market instability. Second, we develop a moment-based quantum algorithm and show that low-order moments, including the relative trace, are strongly correlated with high-dimensional Betti information, even when the relative Betti number is small. Finally, we present circuit constructions, resource estimates, quantum-classical crossover projections, and experimental results from a Barium development system similar to the forthcoming IonQ Tempo line, extracting Laplacian-derived observables from graph instances and quantitatively comparing them with exact Betti information. Together, these results establish quantum TDA as a practical approach for extracting topological features from classically challenging data