Featured Papers

Main idea

Adapts a quantum convolutional neural network into a quantum autoencoder for reconstruction-based anomaly detection, trained semi-supervised on normal samples with reconstruction error as the anomaly score. It compares two designs — a hierarchical architecture that keeps information distributed across the circuit, and a bottleneck architecture that explicitly compresses information into a smaller latent space — benchmarked against a variational quantum circuit and a classical baseline on real exoplanet data.

Main idea

Demonstrates nine quantum error-correcting codes with very different connectivity requirements on a single trapped-ion device without hardware reconfiguration, spanning qLDPC, topological, and concatenated codes. A qLDPC code encoding 4 logical qubits into 18 physical qubits achieves a logical error rate up to 9x better than a prior superconducting demonstration — and reaches breakeven, where some logical qubit lifetimes meet or exceed the physical qubits.

Main idea

Presents a qubit-efficient encoding for Kohn-Sham density functional theory wavefunctions, with a quantum algorithm that computes all occupied orbitals simultaneously. By avoiding the costly readout of the electronic density — a key bottleneck — the method enables total energy evaluation with a potential exponential speedup for the Harris functional.

Main idea

Adapts Microsoft's QuantumKatas curriculum from Q# to Qiskit and packages it as a benchmark of 350 tasks across 26 categories with deterministic simulation-based verification. Evaluating 16 LLMs across 7 prompting setups (39,200 runs), the benchmark cleanly separates models with pass rates from 32% to 83%.

Main idea

A multi-task Soft Actor-Critic reinforcement learning framework for open-system quantum control that learns optimal pulse sequences while simultaneously discovering the problem-specific evolution time T and number of control pulse segments N. Across 51 Hamiltonian variations under environmental noise, it drives systems from initial to target states with high fidelity, and tests whether a single model generalizes to Hamiltonians not seen during training.

Main idea

Uses dissipative state preparation on IBM quantum processors to prepare approximate thermal states of frustrated spin models on the kagome lattice — including the antiferromagnetic Heisenberg model, whose finite-temperature properties are inaccessible to quantum Monte Carlo due to a severe sign problem. They reach 79 spins coupled to 60 environment qubits (139 total), with a robust steady state of adjustable effective temperature persisting through over 1,000 layers of two-qubit gates.

Main idea

Identifies a sharp interaction-degree threshold for classically simulating QAOA with 2-local cost functions. At degree 2, exact classical sampling from depth-p QAOA runs in polynomial time when p = O(log n). At degree 3, classical sampling with even small multiplicative error would collapse the polynomial hierarchy to its third level — strong evidence of genuine hardness.

Main idea

Inserts Cayley-parameterised unitary adapters — quantum circuit blocks — into the frozen projection layers of pre-trained LLMs, executed on a 156-qubit IBM Quantum System Two. The method improves the perplexity of Llama 3.1 8B by 1.4% with only 6,000 additional parameters, with end-to-end inference validated on real quantum hardware, not a simulator.

Main idea

Provides new resource estimates for breaking the 256-bit Elliptic Curve Discrete Logarithm Problem — the core of modern blockchain cryptography. Shor's algorithm can run with under 1,450 logical qubits and under 70 million Toffoli gates; on superconducting hardware with 1e-3 error rates, the attack executes in minutes using fewer than half a million physical qubits. Introduces a fast-clock vs slow-clock architecture distinction that determines whether real-time on-spend attacks are feasible.

Main idea

Google Willow processor demonstrates two surface code memories operating below the error correction threshold. Increasing the code distance from 5 to 7 reduces the logical error rate by a factor of 2.14. The logical qubit lifetime exceeds the best physical qubit by 2.4x.

Main idea

Demonstrates fault-tolerant quantum memory using low-density parity-check (LDPC) codes achieving an error threshold of 0.7% on par with surface codes but with dramatically lower qubit overhead. LDPC codes can encode more logical qubits per physical qubit than surface codes.

Main idea

Proves that specific circuit architectures with finite local depth and long-range entanglement can avoid barren plateaus even at scale. Provides a constructive route to trainable deep quantum circuits.

Main idea

Proves a direct mathematical connection between circuit expressibility and barren plateaus: circuits that can approximate 2-designs (highly expressive) necessarily have exponentially vanishing gradients. The trainability-expressibility trade-off is fundamental and inescapable.

Main idea

A comprehensive survey of variational quantum algorithms — covering design, training, applications to chemistry, optimization, and ML, alongside the key challenges: barren plateaus, noise, and the difficulty of demonstrating quantum advantage.

Main idea

Provides a rigorous proof-of-principle quantum advantage in supervised ML for a specific problem based on discrete logarithm — a problem believed to be hard classically but easy quantumly. The dataset was artificially constructed to demonstrate the separation.

Main idea

Quantum ML models do not automatically outperform classical ones. By carefully constructing classical kernels that mimic quantum feature maps, classical ML can match or exceed quantum models on most practical datasets. The key insight is that quantum advantage in ML requires data with genuine quantum structure.

Main idea

A comprehensive resource estimation for fault-tolerant quantum simulation of industrially relevant chemistry problems. Shows that simulating molecules like FeMoco for nitrogen fixation requires millions of T gates and thousands of logical qubits.

Main idea

A detailed resource estimation for running Shor's algorithm at practical scale. Breaking RSA-2048 requires approximately 20 million physical qubits running for 8 hours with surface code error correction at realistic error rates.

Main idea

Proves that all supervised quantum ML models are equivalent to kernel methods, where the kernel is defined by the inner product of quantum feature states. This unifies QML with classical kernel theory and provides rigorous tools for analyzing quantum models.

Main idea

Barren plateaus depend critically on whether the cost function is global or local. Local cost functions on shallow circuits can avoid barren plateaus even at scale, providing a partial path forward for trainable VQAs.