-
Meta-Designing Quantum Experiments with Language Models
Authors:
Sören Arlt,
Haonan Duan,
Felix Li,
Sang Michael Xie,
Yuhuai Wu,
Mario Krenn
Abstract:
Artificial Intelligence (AI) has the potential to significantly advance scientific discovery by finding solutions beyond human capabilities. However, these super-human solutions are often unintuitive and require considerable effort to uncover underlying principles, if possible at all. Here, we show how a code-generating language model trained on synthetic data can not only find solutions to specif…
▽ More
Artificial Intelligence (AI) has the potential to significantly advance scientific discovery by finding solutions beyond human capabilities. However, these super-human solutions are often unintuitive and require considerable effort to uncover underlying principles, if possible at all. Here, we show how a code-generating language model trained on synthetic data can not only find solutions to specific problems but can create meta-solutions, which solve an entire class of problems in one shot and simultaneously offer insight into the underlying design principles. Specifically, for the design of new quantum physics experiments, our sequence-to-sequence transformer architecture generates interpretable Python code that describes experimental blueprints for a whole class of quantum systems. We discover general and previously unknown design rules for infinitely large classes of quantum states. The ability to automatically generate generalized patterns in readable computer code is a crucial step toward machines that help discover new scientific understanding -- one of the central aims of physics.
△ Less
Submitted 4 June, 2024;
originally announced June 2024.
-
Dismantling Common Internet Services for Ad-Malware Detection
Authors:
Florian Nettersheim,
Stephan Arlt,
Michael Rademacher
Abstract:
Online advertising represents a main instrument for publishers to fund content on the World Wide Web. Unfortunately, a significant number of online advertisements often accommodates potentially malicious content, such as cryptojacking hidden in web banners - even on reputable websites. In order to protect Internet users from such online threats, the thorough detection of ad-malware campaigns plays…
▽ More
Online advertising represents a main instrument for publishers to fund content on the World Wide Web. Unfortunately, a significant number of online advertisements often accommodates potentially malicious content, such as cryptojacking hidden in web banners - even on reputable websites. In order to protect Internet users from such online threats, the thorough detection of ad-malware campaigns plays a crucial role for a safe Web. Today, common Internet services like VirusTotal can label suspicious content based on feedback from contributors and from the entire Web community. However, it is open to which extent ad-malware is actually taken into account and whether the results of these services are consistent. In this pre-study, we evaluate who defines ad-malware on the Internet. In a first step, we crawl a vast set of websites and fetch all HTTP requests (particularly to online advertisements) within these websites. Then we query these requests both against popular filtered DNS providers and VirusTotal. The idea is to validate, how much content is labeled as a potential threat. The results show that up to 0.47% of the domains found during crawling are labeled as suspicious by DNS providers and up to 8.8% by VirusTotal. Moreover, only about 0.7% to 3.2% of these domains are categorized as ad-malware. The overall responses from the used Internet services paint a divergent picture: All considered services have different understandings to the definition of suspicious content. Thus, we outline potential research efforts to the automated detection of ad-malware. We further bring up the open question of a common definition of ad-malware to the Web community.
△ Less
Submitted 22 April, 2024;
originally announced April 2024.
-
Virtual Reality for Understanding Artificial-Intelligence-driven Scientific Discovery with an Application in Quantum Optics
Authors:
Philipp Schmidt,
Sören Arlt,
Carlos Ruiz-Gonzalez,
Xuemei Gu,
Carla Rodríguez,
Mario Krenn
Abstract:
Generative Artificial Intelligence (AI) models can propose solutions to scientific problems beyond human capability. To truly make conceptual contributions, researchers need to be capable of understanding the AI-generated structures and extracting the underlying concepts and ideas. When algorithms provide little explanatory reasoning alongside the output, scientists have to reverse-engineer the fu…
▽ More
Generative Artificial Intelligence (AI) models can propose solutions to scientific problems beyond human capability. To truly make conceptual contributions, researchers need to be capable of understanding the AI-generated structures and extracting the underlying concepts and ideas. When algorithms provide little explanatory reasoning alongside the output, scientists have to reverse-engineer the fundamental insights behind proposals based solely on examples. This task can be challenging as the output is often highly complex and thus not immediately accessible to humans. In this work we show how transferring part of the analysis process into an immersive Virtual Reality (VR) environment can assist researchers in develo** an understanding of AI-generated solutions. We demonstrate the usefulness of VR in finding interpretable configurations of abstract graphs, representing Quantum Optics experiments. Thereby, we can manually discover new generalizations of AI-discoveries as well as new understanding in experimental quantum optics. Furthermore, it allows us to customize the search space in an informed way - as a human-in-the-loop - to achieve significantly faster subsequent discovery iterations. As concrete examples, with this technology, we discover a new resource-efficient 3-dimensional entanglement swap** scheme, as well as a 3-dimensional 4-particle Greenberger-Horne-Zeilinger-state analyzer. Our results show the potential of VR for increasing a human researcher's ability to derive knowledge from graph-based generative AI that, which is a common abstract data representation used in diverse fields of science.
△ Less
Submitted 20 February, 2024;
originally announced March 2024.
-
XLuminA: An Auto-differentiating Discovery Framework for Super-Resolution Microscopy
Authors:
Carla Rodríguez,
Sören Arlt,
Leonhard Möckl,
Mario Krenn
Abstract:
Driven by human ingenuity and creativity, the discovery of super-resolution techniques, which circumvent the classical diffraction limit of light, represent a leap in optical microscopy. However, the vast space encompassing all possible experimental configurations suggests that some powerful concepts and techniques might have not been discovered yet, and might never be with a human-driven direct d…
▽ More
Driven by human ingenuity and creativity, the discovery of super-resolution techniques, which circumvent the classical diffraction limit of light, represent a leap in optical microscopy. However, the vast space encompassing all possible experimental configurations suggests that some powerful concepts and techniques might have not been discovered yet, and might never be with a human-driven direct design approach. Thus, AI-based exploration techniques could provide enormous benefit, by exploring this space in a fast, unbiased way. We introduce XLuminA, an open-source computational framework developed using JAX, which offers enhanced computational speed enabled by its accelerated linear algebra compiler (XLA), just-in-time compilation, and its seamlessly integrated automatic vectorization, auto-differentiation capabilities and GPU compatibility. Remarkably, XLuminA demonstrates a speed-up of 4 orders of magnitude compared to well-established numerical optimization methods. We showcase XLuminA's potential by re-discovering three foundational experiments in advanced microscopy. Ultimately, XLuminA identified a novel experimental blueprint featuring sub-diffraction imaging capabilities. This work constitutes an important step in AI-driven scientific discovery of new concepts in optics and advanced microscopy.
△ Less
Submitted 17 May, 2024; v1 submitted 12 October, 2023;
originally announced October 2023.
-
Deep Quantum Graph Dreaming: Deciphering Neural Network Insights into Quantum Experiments
Authors:
Tareq Jaouni,
Sören Arlt,
Carlos Ruiz-Gonzalez,
Ebrahim Karimi,
Xuemei Gu,
Mario Krenn
Abstract:
Despite their promise to facilitate new scientific discoveries, the opaqueness of neural networks presents a challenge in interpreting the logic behind their findings. Here, we use a eXplainable-AI (XAI) technique called $inception$ or $deep$ $dreaming$, which has been invented in machine learning for computer vision. We use this technique to explore what neural networks learn about quantum optics…
▽ More
Despite their promise to facilitate new scientific discoveries, the opaqueness of neural networks presents a challenge in interpreting the logic behind their findings. Here, we use a eXplainable-AI (XAI) technique called $inception$ or $deep$ $dreaming$, which has been invented in machine learning for computer vision. We use this technique to explore what neural networks learn about quantum optics experiments. Our story begins by training deep neural networks on the properties of quantum systems. Once trained, we "invert" the neural network -- effectively asking how it imagines a quantum system with a specific property, and how it would continuously modify the quantum system to change a property. We find that the network can shift the initial distribution of properties of the quantum system, and we can conceptualize the learned strategies of the neural network. Interestingly, we find that, in the first layers, the neural network identifies simple properties, while in the deeper ones, it can identify complex quantum structures and even quantum entanglement. This is in reminiscence of long-understood properties known in computer vision, which we now identify in a complex natural science task. Our approach could be useful in a more interpretable way to develop new advanced AI-based scientific discovery techniques in quantum physics.
△ Less
Submitted 4 October, 2023; v1 submitted 13 September, 2023;
originally announced September 2023.
-
Experimental Solutions to the High-Dimensional Mean King's Problem
Authors:
Tareq Jaouni,
Xiaoqin Gao,
Sören Arlt,
Mario Krenn,
Ebrahim Karimi
Abstract:
In 1987, Vaidman, Aharanov, and Albert put forward a puzzle called the Mean King's Problem (MKP) that can be solved only by harnessing quantum entanglement. Prime-powered solutions to the problem have been shown to exist, but they have not yet been experimentally realized for any dimension beyond two. We propose a general first-of-its-kind experimental scheme for solving the MKP in prime dimension…
▽ More
In 1987, Vaidman, Aharanov, and Albert put forward a puzzle called the Mean King's Problem (MKP) that can be solved only by harnessing quantum entanglement. Prime-powered solutions to the problem have been shown to exist, but they have not yet been experimentally realized for any dimension beyond two. We propose a general first-of-its-kind experimental scheme for solving the MKP in prime dimensions ($D$). Our search is guided by the digital discovery framework PyTheus, which finds highly interpretable graph-based representations of quantum optical experimental setups; using it, we find specific solutions and generalize to higher dimensions through human insight. As proof of principle, we present a detailed investigation of our solution for the three-, five-, and seven-dimensional cases. We obtain maximum success probabilities of $72.8 \%$, $45.8\%$, and $34.8 \%$, respectively. We, therefore, posit that our computer-inspired scheme yields solutions that exceed the classical probability ($1/D$) twofold, demonstrating its promise for experimental implementation.
△ Less
Submitted 25 July, 2023; v1 submitted 24 July, 2023;
originally announced July 2023.
-
Digital Discovery of a Scientific Concept at the Core of Experimental Quantum Optics
Authors:
Sören Arlt,
Carlos Ruiz-Gonzalez,
Mario Krenn
Abstract:
Entanglement is a crucial resource for quantum technologies ranging from quantum communication to quantum-enhanced measurements and computation. Finding experimental setups for these tasks is a conceptual challenge for human scientists due to the counterintuitive behavior of multiparticle interference and the enormously large combinatorial search space. Recently, new possibilities have been opened…
▽ More
Entanglement is a crucial resource for quantum technologies ranging from quantum communication to quantum-enhanced measurements and computation. Finding experimental setups for these tasks is a conceptual challenge for human scientists due to the counterintuitive behavior of multiparticle interference and the enormously large combinatorial search space. Recently, new possibilities have been opened by artificial discovery where artificial intelligence proposes experimental setups for the creation and manipulation of high-dimensional multi-particle entanglement. While digitally discovered experiments go beyond what has been conceived by human experts, a crucial goal is to understand the underlying concepts which enable these new useful experimental blueprints. Here, we present Halo (Hyperedge Assembly by Linear Optics), a new form of multiphoton quantum interference with surprising properties. Halos were used by our digital discovery framework to solve previously open questions. We -- the human part of this collaboration -- were then able to conceptualize the idea behind the computer discovery and describe them in terms of effective probabilistic multi-photon emitters. We then demonstrate its usefulness as a core of new experiments for highly entangled states, communication in quantum networks, and photonic quantum gates. Our manuscript has two conclusions. First, we introduce and explain the physics of a new practically useful multi-photon interference phenomenon that can readily be realized in advanced setups such as integrated photonic circuits. Second, our manuscript demonstrates how artificial intelligence can act as a source of inspiration for the scientific discoveries of new actionable concepts in physics.
△ Less
Submitted 18 October, 2022;
originally announced October 2022.
-
Digital Discovery of 100 diverse Quantum Experiments with PyTheus
Authors:
Carlos Ruiz-Gonzalez,
Sören Arlt,
Jan Petermann,
Sharareh Sayyad,
Tareq Jaouni,
Ebrahim Karimi,
Nora Tischler,
Xuemei Gu,
Mario Krenn
Abstract:
Photons are the physical system of choice for performing experimental tests of the foundations of quantum mechanics. Furthermore, photonic quantum technology is a main player in the second quantum revolution, promising the development of better sensors, secure communications, and quantum-enhanced computation. These endeavors require generating specific quantum states or efficiently performing quan…
▽ More
Photons are the physical system of choice for performing experimental tests of the foundations of quantum mechanics. Furthermore, photonic quantum technology is a main player in the second quantum revolution, promising the development of better sensors, secure communications, and quantum-enhanced computation. These endeavors require generating specific quantum states or efficiently performing quantum tasks. The design of the corresponding optical experiments was historically powered by human creativity but is recently being automated with advanced computer algorithms and artificial intelligence. While several computer-designed experiments have been experimentally realized, this approach has not yet been widely adopted by the broader photonic quantum optics community. The main roadblocks consist of most systems being closed-source, inefficient, or targeted to very specific use-cases that are difficult to generalize. Here, we overcome these problems with a highly-efficient, open-source digital discovery framework PyTheus, which can employ a wide range of experimental devices from modern quantum labs to solve various tasks. This includes the discovery of highly entangled quantum states, quantum measurement schemes, quantum communication protocols, multi-particle quantum gates, as well as the optimization of continuous and discrete properties of quantum experiments or quantum states. PyTheus produces interpretable designs for complex experimental problems which human researchers can often readily conceptualize. PyTheus is an example of a powerful framework that can lead to scientific discoveries -- one of the core goals of artificial intelligence in science. We hope it will help accelerate the development of quantum optics and provide new ideas in quantum hardware and technology.
△ Less
Submitted 8 December, 2023; v1 submitted 18 October, 2022;
originally announced October 2022.
-
Black-Box Verification for GUI Applications
Authors:
Stephan Arlt,
Evren Ermis,
Sergio Feo-Arenis,
Andreas Podelski
Abstract:
In black-box testing of GUI applications (a form of system testing), a dynamic analysis of the GUI application is used to infer a black-box model; the black-box model is then used to derive test cases for the test of the GUI application. In this paper, we propose to supplement the test with the verification of the black-box model. We present a method that can give a guarantee of the absence of fau…
▽ More
In black-box testing of GUI applications (a form of system testing), a dynamic analysis of the GUI application is used to infer a black-box model; the black-box model is then used to derive test cases for the test of the GUI application. In this paper, we propose to supplement the test with the verification of the black-box model. We present a method that can give a guarantee of the absence of faults, i.e., the correctness of all test cases of the black-box model. The black-model allows us to formulate a parametrized verification problem. As we will show, it also allows us to circumvent the static analysis of the GUI tool kit. We have implemented our approach; preliminary experiments indicate its practical potential.
△ Less
Submitted 17 October, 2012;
originally announced October 2012.
-
Grey-box GUI Testing: Efficient Generation of Event Sequences
Authors:
Stephan Arlt,
Ishan Banerjee,
Cristiano Bertolini,
Atif M. Memon,
Martin Schäf
Abstract:
Graphical user interfaces (GUIs), due to their event driven nature, present a potentially unbounded space of all possible ways to interact with software. During testing it becomes necessary to effectively sample this space. In this paper we develop algorithms that sample the GUI's input space by only generating sequences that (1) are allowed by the GUI's structure, and (2) chain together only thos…
▽ More
Graphical user interfaces (GUIs), due to their event driven nature, present a potentially unbounded space of all possible ways to interact with software. During testing it becomes necessary to effectively sample this space. In this paper we develop algorithms that sample the GUI's input space by only generating sequences that (1) are allowed by the GUI's structure, and (2) chain together only those events that have data dependencies between their event handlers. We create a new abstraction, called an event-dependency graph (EDG) of the GUI, that captures data dependencies between event handler code. We develop a map** between EDGs and an existing black-box user-level model of the GUI's workflow, called an event-flow graph (EFG). We have implemented automated EDG construction in a tool that analyzes the bytecode of each event handler. We evaluate our "grey-box" approach using four open-source applications and compare it with the current state-of-the-art EFG approach. Our results show that using the EDG reduces the number of test cases while still achieving at least the same coverage. Furthermore, we were able to detect 2 new bugs in the subject applications.
△ Less
Submitted 22 May, 2012;
originally announced May 2012.
