Abstract:Many problems in high-dimensional statistics appear to have a statistical-computational gap: a range of values of the signal-to-noise ratio where inference is information-theoretically possible, but (conjecturally) computationally intractable. A canonical such problem is Tensor PCA, where we observe a tensor $Y$ consisting of a rank-one signal plus Gaussian noise. Multiple lines of work suggest that Tensor PCA becomes computationally hard at a critical value of the signal's magnitude. In particular, below this transition, no low-degree polynomial algorithm can detect the signal with high probability; conversely, various spectral algorithms are known to succeed above this transition. We unify and extend this work by considering tensor networks, orthogonally invariant polynomials where multiple copies of $Y$ are "contracted" to produce scalars, vectors, matrices, or other tensors. We define a new set of objects, tensor cumulants, which provide an explicit, near-orthogonal basis for invariant polynomials of a given degree. This basis lets us unify and strengthen previous results on low-degree hardness, giving a combinatorial explanation of the hardness transition and of a continuum of subexponential-time algorithms that work below it, and proving tight lower bounds against low-degree polynomials for recovering rather than just detecting the signal. It also lets us analyze a new problem of distinguishing between different tensor ensembles, such as Wigner and Wishart tensors, establishing a sharp computational threshold and giving evidence of a new statistical-computational gap in the Central Limit Theorem for random tensors. Finally, we believe these cumulants are valuable mathematical objects in their own right: they generalize the free cumulants of free probability theory from matrices to tensors, and share many of their properties, including additivity under additive free convolution.

Tensor Clustering with Planted Structures: Statistical Optimality and Computational Limits

Exact Clustering in Tensor Block Model: Statistical Optimality and Computational Limit

Heteroskedastic Tensor Clustering

Exact Partitioning of High-order Planted Models with a Tensor Nuclear Norm Constraint

Robust Subspace Clustering Via Thresholding Ridge Regression

Robust Tensor Clustering with Non-Greedy Maximization.

Tensor clustering with algebraic constraints gives interpretable groups of crosstalk mechanisms in breast cancer

Tensor cumulants for statistical inference on invariant distributions

Tensor Multi-Clustering Parallel Intelligent Computing Method Based on Tensor Chain Decomposition

Scalable tensor methods for nonuniform hypergraphs

Breaking the Small Cluster Barrier of Graph Clustering

Testing Higher-order Clusterability on graphs

Jointly Modeling and Clustering Tensors in High Dimensions

(Un)detectable cluster structure in sparse networks

Tensor Low-Rank Representation for Data Recovery and Clustering

Tensor LRR and Sparse Coding-Based Subspace Clustering.

Tensor Double Arc-tangent Norm Minimization for Multi-view Clustering

Sparse clusterability: testing for cluster structure in high dimensions

Tensor Graphical Model: Non-Convex Optimization and Statistical Inference

Robust spectral clustering with rank statistics

Detecting cluster patterns in tensor data