arXiv subjects

Francisco Ferreira da Silva

Publications and source records attributed to Francisco Ferreira da Silva.

11 recordsLinked to original sources

arXiv2026

Compressed Computation under $L^4$ Loss is likely Computation in Superposition

Neural networks are thought to represent concepts as directions in their activation space, and superposition lets them encode more concepts than they have dimensions. It is natural to ask whether they can also compute more functions than they have neurons, i.e., perform computation in superposition. In this regime many functions of sparse inputs are evaluated by a layer with fewer neurons than there are functions to compute. Representation in superposition is by now fairly well understood, but computation in superposition is not, and there are few toy models of it arising through training rather than being hand designed. As a toy model of computation in superposition we study the compressed-computation setup: a single-hidden-layer ReLU network with 50 neurons that must compute the ReLU of each of 100 sparse input features. We show that training it under an $L^4$ loss (the mean fourth power of the error), rather than the usual $L^2$, elicits a solution that appears to compute all features in superposition. We then reverse-engineer this solution. We find that the network assigns each feature a sparse binary codeword over neurons and decodes it with a pseudoinverse of the encoder. Given these codewords, a description with only three scalars recovers most of the network's performance, and we validate it by building equivalent networks from hand-designed codes.

cs.LG↗

arXiv2026

Evidence for feature-specific error correction in LLMs

Understanding the features of large language models (LLMs) is a central goal of interpretability. LLMs are commonly assumed to use superposition to represent more features than they have dimensions. They may not only represent features in superposition but also perform computation in superposition. Theory predicts that computing in superposition requires error correction that privileges feature directions over generic ones, but this prediction has not been tested empirically. We propose an empirical test of error correction in LLMs based on activation perturbations. Perturbing residual-stream activations, we find that they are robust to small perturbations--forming activation plateaus consistent with error correction--but less robust along candidate feature directions ("pure" directions, constructed from contrastive prompt pairs) than along mixtures of two such directions, indicating that the pure directions are privileged. We quantify this privilegedness by modeling the perturbation effect as a function of the $L^p$-norm of its decomposition into feature components. For $p=2$ the response is a quadratic form with at most as many nonzero eigenvalues as the residual-stream dimension, which cannot privilege the many feature directions superposition requires. $p>2$ lifts this constraint and is consistent with feature-specific error correction. We find $p>2$ for contrastive, MELBO, and SAE-decoder directions, and $p\approx2$ for random and PCA directions (controls). These results replicate across Gemma-2-9B, Qwen3-1.7B, Llama-3.1-8B, Mistral-7B-v0.3, Aya-Expanse-8B, and Yi-1.5-9B. We further validate our method on a toy model of error correction with known ground-truth features, recovering $p>2$ for true feature directions, degrading toward $2$ as we rotate away from them.

cs.LG↗

arXiv2026

Entanglement improves coordination in distributed systems

Coordination in distributed systems is often hampered by communication latency, which degrades performance. Quantum entanglement offers fundamentally stronger correlations than classically achievable without communication. Crucially, these correlations manifest instantaneously upon measurement, irrespective of the physical distance separating the systems. We investigate the application of shared entanglement to a dual-work optimization problem in a distributed system comprising two servers. The system must process both a continuously available, preemptible baseline task and incoming customer requests arriving in pairs. System performance is characterized by the trade-off between baseline task throughput and customer waiting time. We present a rigorous analytical model demonstrating that when the baseline task throughput function is strictly convex, rewarding longer uninterrupted processing periods, entanglement-assisted routing strategies achieve Pareto-superior performance compared to optimal communication-free classical strategies. We prove this advantage through queueing-theoretic analysis, non-local game formulation, and computational certification of classical bounds. Our results identify distributed scheduling and coordination as a novel application domain for near-term entanglement-based quantum networks.

quant-ph↗

arXiv2026

Reinforcement Learning for Quantum Network Control with Application-Driven Objectives

Optimized control of quantum networks is essential for enabling distributed quantum applications with strict performance requirements. In near-term architectures with constrained hardware, effective control may determine the feasibility of deploying such applications. Because quantum network dynamics are suitable for being modeled as a Markov decision process, dynamic programming and reinforcement learning (RL) offer promising tools for optimizing control strategies. However, key quantum network performance measures -- such as secret key rate in quantum key distribution -- often involve a non-linear relationship between interdependent variables that describe quantum state quality and generation rate. Such objectives are not easily captured by standard RL approaches based on additive rewards. We propose a novel gradient-based RL framework that directly optimizes non-linear, differentiable objective functions, while accounting for uncertainties introduced by classical communication delays. We evaluate this framework in the context of entanglement distillation between two quantum network nodes equipped with multiplexing capability, and demonstrate up to 20-23% improvement over heuristic baselines in certain parameter regimes. Our work comprises the first step towards non-linear objective function optimization in quantum networks with RL, opening a path towards more advanced use cases.

quant-ph↗

arXiv2024

Hardware requirements for trapped-ion based verifiable blind quantum computing with a measurement-only client

In blind quantum computing, a user with a simple client device can perform a quantum computation on a remote quantum server such that the server cannot gain knowledge about the computation. Here, we numerically investigate hardware requirements for verifiable blind quantum computing using an ion trap as server and a distant measurement-only client. While the client has no direct access to quantum-computing resources, it can remotely execute quantum programs on the server by measuring photons emitted by the trapped ion. We introduce a numerical model for trapped-ion quantum devices in NetSquid, a discrete-event simulator for quantum networks. Using this, we determine the minimal hardware requirements on a per-parameter basis to perform the verifiable blind quantum computing protocol. We benchmark these for a five-qubit linear graph state, with which any single-qubit rotation can be performed, where client and server are separated by 50 km. Current state-of-the-art ion traps satisfy the minimal requirements on a per-parameter basis, but all current imperfections combined make it impossible to perform the blind computation securely over 50 km using existing technology. Using a genetic algorithm, we determine the set of hardware parameters that minimises the total improvements required, finding directions along which to improve hardware to reach our threshold error probability that would enable experimental demonstration. In this way, we lay a path for the near-term experimental progress required to realise the implementation of verifiable blind quantum computing over a 50 km distance.

quant-ph↗

arXiv2023

Requirements for a processing-node quantum repeater on a real-world fiber grid

We numerically study the distribution of entanglement between the Dutch cities of Delft and Eindhoven realized with a processing-node quantum repeater and determine minimal hardware requirements for verifiable blind quantum computation using color centers and trapped ions. Our results are obtained considering restrictions imposed by a real-world fiber grid and using detailed hardware-specific models. By comparing our results to those we would obtain in idealized settings we show that simplifications lead to a distorted picture of hardware demands, particularly on memory coherence and photon collection. We develop general machinery suitable for studying arbitrary processing-node repeater chains using NetSquid, a discrete-event simulator for quantum networks. This enables us to include time-dependent noise models and simulate repeater protocols with cut-offs, including the required classical control communication. We find minimal hardware requirements by solving an optimization problem using genetic algorithms on a high-performance-computing cluster. Our work provides guidance for further experimental progress, and showcases limitations of studying quantum-repeater requirements in idealized situations.

quant-ph↗

arXiv2023

Reducing hardware requirements for entanglement distribution via joint hardware-protocol optimization

We conduct a numerical investigation of fiber-based entanglement distribution over distances of up to 1600km using a chain of processing-node quantum repeaters. We determine minimal hardware requirements while simultaneously optimizing over protocols for entanglement generation and entanglement purification, as well as over strategies for entanglement swapping. Notably, we discover that through an adequate choice of protocols the hardware improvement cost scales linearly with the distance covered. Our results highlight the crucial role of good protocol choices in significantly reducing hardware requirements, such as employing purification to meet high-fidelity targets and adopting a SWAP-ASAP policy for faster rates. To carry out this analysis, we employ an extensive simulation framework implemented with NetSquid, a discrete-event-based quantum-network simulator, and a genetic-algorithm-based optimization methodology to determine minimal hardware requirements.

quant-ph↗

arXiv2023

Requirements for upgrading trusted nodes to a repeater chain over 900 km of optical fiber

We perform a numerical study of the distribution of entanglement on a real-world fiber grid connecting the German cities of Bonn and Berlin. The connection is realized using a chain of processing-node quantum repeaters spanning roughly 900 kilometers. We investigate how minimal hardware requirements depend on the target application, as well as on the number of repeaters in the chain. We find that requirements for blind quantum computing are markedly different than those for quantum key distribution, with the required coherence time being around two and a half times larger for the former. Further, we observe a trade-off regarding how target secret-key rates are achieved when using different numbers of repeaters: comparatively low-quality entangled states generated at a high rate are preferred for higher numbers of repeaters, whereas comparatively high-quality states generated at a lower rate are favored for lower numbers of repeaters. To obtain our results we employ an extensive simulation framework implemented using NetSquid, a discrete-event simulator for quantum networks. These are combined with an optimization methodology based on genetic algorithms to determine minimal hardware requirements.

quant-ph↗

arXiv2022

Quantum dynamics for energetic advantage in a charge-based classical full-adder

We present a proposal for a one-bit full-adder to process classical information based on the quantum reversible dynamics of a triple quantum dot system. The device works via the repeated execution of a Fredkin gate implemented through the dynamics of a single time-independent Hamiltonian. Our proposal uses realistic parameter values and could be implemented on currently available quantum dot architectures. We compare the estimated energy requirements for operating our full-adder with those of well-known fully classical devices, and argue that our proposal may provide a consistently better energy efficiency. Our work serves as a proof of principle for the development of energy-efficient information technologies operating through coherent quantum dynamics.

quant-ph↗

arXiv2022

Benchmarking of Quantum Protocols

Quantum network protocols offer new functionalities such as enhanced security to communication and computational systems. Despite the rapid progress in quantum hardware, it has not yet reached a level of maturity that enables execution of many quantum protocols in practical settings. To develop quantum protocols in real world, it is necessary to examine their performance considering the imperfections in their practical implementation using simulation platforms. In this paper, we consider several quantum protocols that enable promising functionalities and services in near-future quantum networks. The protocols are chosen from both areas of quantum communication and quantum computation as follows: quantum money, W-state based anonymous transmission, verifiable blind quantum computation, and quantum digital signature. We use NetSquid simulation platform to evaluate the effect of various sources of noise on the performance of these protocols, considering different figures of merit. We find that to enable quantum money protocol, the decoherence time constant of the quantum memory must be at least three times the storage time of qubits. Furthermore, our simulation results for the w-state based anonymous transmission protocol show that to achieve an average fidelity above 0.8 in this protocol, the storage time of sender's and receiver's particles in the quantum memory must be less than half of the decoherence time constant of the quantum memory. We have also investigated the effect of gate imperfections on the performance of verifiable blind quantum computation. We find that with our chosen parameters, if the depolarizing probability of quantum gates is equal to or greater than 0.05, the security of the protocol cannot be guaranteed. Lastly, our simulation results for quantum digital signature protocol show that channel loss has a significant effect on the probability of repudiation.

quant-ph↗

arXiv2020

Optimizing Entanglement Generation and Distribution Using Genetic Algorithms

Long-distance quantum communication via entanglement distribution is of great importance for the quantum internet. However, scaling up to such long distances has proved challenging due to the loss of photons, which grows exponentially with the distance covered. Quantum repeaters could in theory be used to extend the distances over which entanglement can be distributed, but in practice hardware quality is still lacking. Furthermore, it is generally not clear how an improvement in a certain repeater parameter, such as memory quality or attempt rate, impacts the overall network performance, rendering the path towards scalable quantum repeaters unclear. In this work we propose a methodology based on genetic algorithms and simulations of quantum repeater chains for optimization of entanglement generation and distribution. By applying it to simulations of several different repeater chains, including real-world fiber topology, we demonstrate that it can be used to answer questions such as what are the minimum viable quantum repeaters satisfying given network performance benchmarks. This methodology constitutes an invaluable tool for the development of a blueprint for a pan-European quantum internet. We have made our code, in the form of NetSquid simulations and the smart-stopos optimization tool, freely available for use either locally or on high-performance computing centers.

quant-ph↗