6arXiv cs.LG (Machine Learning)·
Provable quantum-classical learning separation for predicting quantum many-body time evolution
A new arXiv preprint establishes a rigorous PAC-learning separation between quantum and classical algorithms for predicting the time evolution of quantum many-body systems under an unknown Hamiltonian. The quantum procedure learns the Hamiltonian from short-time samples and uses Hamiltonian simulation with classical shadows for inference, while classical hardness is proven by embedding a BQP-complete computation into Feynman-Kitaev clock Hamiltonian dynamics, ruling out classical polynomial-time solutions unless BQP ⊆ P/poly. The result connects quantum learning theory, quantum simulation, and quantum machine learning, providing one of the first provable learning separations for a physically motivated QML task.
Related guides (1)
Related events (8)
Empirical comparison finds quantum ML models do not yet surpass classical baselines
A new arXiv preprint presents a systematic empirical comparison of seven quantum machine learning (QML) model pairs against classical counterparts across supervised learning and reinforcement learning tasks. Results show QML models do not yet surpass classical baselines in prediction performance, policy stability, or training time, though some promise is noted for noise filtering and false positive control. The study identifies open challenges in hardware environments, training efficiency, and convergence stability, and releases code publicly.
IA-VQC-DPC: Intervention-aware quantum predictive control with safety attribution for learned policies
A new arXiv preprint introduces Intervention-Aware Variational Quantum Differentiable Predictive Control (IA-VQC-DPC), a framework that trains variational quantum circuit policies under a primal-dual intervention budget to penalize over-reliance on downstream safety filters (Control-Barrier-Function projections). The work also proposes a safety-attribution protocol that decomposes trajectory corrections into policy-level versus filter-level contributions, enabling measurement of whether a policy has genuinely learned safe behavior or is merely being silently repaired by its safety layer. Experiments on BOPTEST building-control emulators show the quantum policy achieves significantly lower pre-filter violations than a matched classical policy at equal parameter budget, with a notable negative result: a learned energy head is only safe when paired with a distribution-aware runtime guard.
Proper positive-only PAC learning characterized via uniform exterior separability
A new arXiv paper settles a long-open question in learning theory: a concept class is properly learnable from positive-only samples if and only if it has finite VC dimension and satisfies a new combinatorial condition called uniform exterior separability. The work reveals surprising separations absent in standard PAC learning — proper vs. improper, randomized vs. deterministic, and cases where no ERM is a valid learner. New combinatorial dimensions introduced may have broader relevance to learning theory.
PAC-Bayesian finite-sample certificates for quadratic closed-loop control via System Level Synthesis
A new arXiv preprint derives PAC-Bayesian bounds for learning-based linear control with quadratic trajectory costs, a setting previously difficult due to unbounded and non-Lipschitz losses. The authors use System Level Synthesis (SLS) parameterization to expose the closed-loop trajectory map, enabling explicit certification and a tractable quadratic upper bound via closed-loop sensitivity quantities. A data-driven bound is provided that doubles as a learning algorithm for control selection, demonstrated on a double integrator where it acts as a sensitivity-aware finite-sample regularizer in the low-data regime. The work bridges statistical learning theory and control theory but is narrow in scope relative to mainstream ML/AI concerns.
LLM-guided search framework discovers new quantum LDPC code families via structured concept evolution
Researchers introduce Structured Concept Evolution (SCE), a framework pairing an LLM with an algebraic mutation grammar to discover lifted-product quantum LDPC code families. The system evolves structured algebraic specifications rather than asking the LLM to design codes from scratch, enabling discovery of both abelian and non-abelian code families competitive with standard designs like bivariate-bicycle codes. Results are achieved using lightweight models (GPT-5.4-mini and GPT-5.4-nano), suggesting LLM-guided combinatorial search can be effective for hard discrete design problems in quantum error correction.
New paper exposes behavioral divergence in quantized LLMs beyond accuracy/perplexity metrics
A new arXiv preprint argues that standard accuracy and perplexity metrics fail to capture behavioral changes induced by post-training quantization. The authors introduce 'correctness agreement', a decision-level metric measuring overlap in correct predictions between base and quantized models, and find behavioral divergence emerges even when task performance appears preserved. Analysis of attention weight distortions reveals non-linear breakpoints at low bit-widths and that query/key projections are more sensitive to quantization than value/output projections. The findings challenge the assumption that quantized models are behaviorally equivalent to their base counterparts.
PAC-Bayes analysis establishes formal expressivity and alignment floors for prompt-conditioned LLMs
A new arXiv preprint models user-LLM interaction as a bilevel cheap-talk game and derives PAC-Bayes bounds showing two irreducible limitations: an 'expressivity floor' where language's finite channel capacity makes distinct tasks indistinguishable, and an 'objective-misalignment floor' where alignment constraints prevent reaching user-ideal outputs. The authors prove that prompt-conditioned LLMs cannot be universal problem solvers, as correct behavior on certain task families is provably unattainable even with infinite data, optimal training, or model scaling. The work suggests multimodal inputs and external memory as potential mitigations by increasing task-relevant information bandwidth.
Q-DIBA: First input-aware dynamic backdoor attack against Quantum Neural Networks
Researchers introduce Q-DIBA, the first input-aware dynamic backdoor attack targeting Quantum Neural Networks (QNNs), addressing limitations of prior fixed-trigger quantum backdoor methods. The approach jointly trains a classical trigger generator and a victim QNN using a three-mode mini-batch strategy and an ensemble density contrastive loss operating on post-ansatz quantum states before measurement. Experiments on MNIST and Fashion-MNIST demonstrate high attack success rates, stealthiness, and resilience against defenses including spectral-signature detection and fine-tuning. The work highlights a novel security threat relevant to near-term quantum machine learning deployments.