arXiv subjects

Thomas Fischbacher

Publications and source records attributed to Thomas Fischbacher.

At least 19 recordsLinked to original sources

arXiv2025

Towards a future space-based, highly scalable AI infrastructure system design

If AI is a foundational general-purpose technology, we should anticipate that demand for AI compute -- and energy -- will continue to grow. The Sun is by far the largest energy source in our solar system, and thus it warrants consideration how future AI infrastructure could most efficiently tap into that power. This work explores a scalable compute system for machine learning in space, using fleets of satellites equipped with solar arrays, inter-satellite links using free-space optics, and Google tensor processing unit (TPU) accelerator chips. To facilitate high-bandwidth, low-latency inter-satellite communication, the satellites would be flown in close proximity. We illustrate the basic approach to formation flight via an 81-satellite cluster of 1 km radius, and describe an approach for using high-precision ML-based models to control large-scale constellations. Trillium TPUs are radiation tested. They survive a total ionizing dose equivalent to a 5 year mission life without permanent failures, and are characterized for bit-flip errors. Launch costs are a critical part of overall system cost; a learning curve analysis suggests launch to low-Earth orbit (LEO) may reach $\lesssim$\$200/kg by the mid-2030s.

cs.DC↗

arXiv2025

Life at the Boundary of Chemical Kinetics and Program Execution

This work introduces a generic quantitative framework for studying dynamical processes that involve interactions of polymer sequences. Possible applications range from quantitative studies of the reaction kinetics of polymerization processes to explorations of the behavior of chemical implementations of computational -- including basic life-like -- processes. This way, we establish a bridge between thermodynamic and computational aspects of systems that are defined in terms of sequence interactions. As by-products of these investigations, we clarify some common confusion around the notion of ``autocatalysis'' and show quantitatively how a chemically implemented Turing machine can operate close to the Landauer bound. Using a Markov process model of polymer sequence composition and dynamical evolution of the Markov process's parameters via an ordinary differential equation (ODE) that arises when taking the double ``chemical'' many-particle limit as well as ``rarefied interactions'' limit, this approach enables -- for example -- accurate quantitative explorations of entropy generation in systems where computation is driven by relaxation to thermodynamic equilibrium. The computational framework internally utilizes the Scheme programming language's intrinsic continuation mechanisms to provide nondeterministic evaluation primitives that allow the user to specify example systems in straight purely functional code, making exploration of all possible relevant sequence composition constellations -- which would be otherwise tedious to write code for -- automatic and hidden from the user. A collection of fully worked out examples elucidate how this modeling approach is quantitatively related to both exact and approximate analytic approaches. These examples can also serve as starting points for further explorations.

cond-mat.dis-nn↗

arXiv2023

Bit-Twiddling Hacks for Gamma Matrices

For some research questions that involve Spin(p, q) representation theory, using symbolic algebra based techniques might be an attractive option for simplifying and manipulating expressions. Yet, for some such problems, especially as they arise in the study of various limits of M-theory (such as dimensional reductions), the complexity of the resulting expressions can become computationally challenging when using popular symbolic algebra packages in a straightforward manner. This work discusses some general properties of Gamma matrices that are computationally useful, down to the level of what one would call "bit-twiddling hacks" in computer science. It is presented in a self-contained way that should be accessible to both physicists and computer scientists. Code is available alongside the TeX source of the preprint version of this article on arXiv.

hep-th↗

arXiv2023

On backpropagating Hessians through ODEs

We discuss the problem of numerically backpropagating Hessians through ordinary differential equations (ODEs) in various contexts and elucidate how different approaches may be favourable in specific situations. We discuss both theoretical and pragmatic aspects such as, respectively, bounds on computational effort and typical impact of framework overhead. Focusing on the approach of hand-implemented ODE-backpropagation, we develop the computation for the Hessian of orbit-nonclosure for a mechanical system. We also clarify the mathematical framework for extending the backward-ODE-evolution of the costate-equation to Hessians, in its most generic form. Some calculations, such as that of the Hessian for orbit non-closure, are performed in a language, defined in terms of a formal grammar, that we introduce to facilitate the tracking of intermediate quantities. As pedagogical examples, we discuss the Hessian of orbit-nonclosure for the higher dimensional harmonic oscillator and conceptually related problems in Newtonian gravitational theory. In particular, applying our approach to the figure-8 three-body orbit, we readily rediscover a distorted-figure-8 solution originally described by Simó. Possible applications may include: improvements to training of `neural ODE'- type deep learning with second-order methods, numerical analysis of quantum corrections around classical paths, and, more broadly, studying options for adjusting an ODE's initial configuration such that the impact on some given objective function is small.

math.OC↗

arXiv2022

New N=1 AdS$_4$ solutions of type IIB supergravity

We construct analytically a new family of supersymmetric AdS$_4$ solutions of IIB supergravity, with the internal space provided by a deformed $S^5\times S^1$. The solutions preserve N=1 supersymmetry and an SO(3) subgroup of isometries of $S^5$, which is broken to U(1) along a flat direction. They are further parametrised by a winding number and a choice of SL(2) duality twist along the circle in an elliptic conjugacy class, thus including both globally geometric and S-fold configurations. We identify these solutions by first constructing a new family of vacua of D=4, $U(4)\ltimes\mathbb{R}^{12}$ gauged maximal supergravity and use exceptional field theory to perform the uplift to ten dimensions. We discuss the relevance of D=5 Wilson loops associated to preserved and broken gauge symmetries in the construction of these classes of solutions.

hep-th↗

arXiv2022

Vacua of $ω$-deformed SO(8) supergravity

We perform a detailed analysis of the vacua of $ω$-deformed SO(8) supergravity in four dimensions. In particular, using Tensorflow-based numerical methods, we track how the equilibria of the theory change when varying the electric-magnetic deformation parameter $ω$. Apart from describing various properties of different equilibria (390 in total), we show that as $ω$ is deformed, the SO(3), N=1 vacuum of the de Wit-Nicolai theory becomes equivalent to a critical point in $U(4)\ltimes\mathbb{R}^{12}$ gauged supergravity with a known uplift in IIB supergravity. The procedure employed here to obtain a new gauging with a guaranteed equilibrium is generic and allows one to obtain further admissible noncompact gaugings via $ω$-deformation, all of which have guaranteed critical points, and some of which may be novel upliftable solutions.

hep-th↗

arXiv2021

From Binary Error Correcting Codes to a Relation Between Maximal D=4 and D=3 Supergravities

This short note provides (TensorFlow-based) numerical evidence for the embedability (in the limit of a scalar parameter going to infinity) of the scalar potential of dyonic N=8, D=4 SO(8) supergravity into the scalar potential of N=16, D=3 SO(8)xSO(8) supergravity. One finds that the dyonic $ω$-angle gets identified with the compact U(1) part of the SL(2) factor of the SL(2)xE7(7) subgroup of E8(8).

hep-th↗

arXiv2021

Single-Photon Image Classification

Quantum computing-based machine learning mainly focuses on quantum computing hardware that is experimentally challenging to realize due to requiring quantum gates that operate at very low temperature. Instead, we demonstrate the existence of a lower performance and much lower effort island on the accuracy-vs-qubits graph that may well be experimentally accessible with room temperature optics. This high temperature "quantum computing toy model" is nevertheless interesting to study as it allows rather accessible explanations of key concepts in quantum computing, in particular interference, entanglement, and the measurement process. We specifically study the problem of classifying an example from the MNIST and Fashion-MNIST datasets, subject to the constraint that we have to make a prediction after the detection of the very first photon that passed a coherently illuminated filter showing the example. Whereas a classical set-up in which a photon is detected after falling on one of the $28\times 28$ image pixels is limited to a (maximum likelihood estimation) accuracy of $21.27\%$ for MNIST, respectively $18.27\%$ for Fashion-MNIST, we show that the theoretically achievable accuracy when exploiting inference by optically transforming the quantum state of the photon is at least $41.27\%$ for MNIST, respectively $36.14\%$ for Fashion-MNIST. We show in detail how to train the corresponding transformation with TensorFlow and also explain how this example can serve as a teaching tool for the measurement process in quantum mechanics.

cs.LG↗

arXiv2020

New AdS$_4$ Vacua in Dyonic ISO(7) Gauged Supergravity

We identify 219 AdS$_4$ solutions in four-dimensional dyonically gauged ISO(7) $\mathcal{N}=8$ supergravity and present some of their properties. One of the new solutions preserves $\mathcal{N}=1$ supersymmetry and provides a rare explicit example of an AdS$_4$ vacuum dual to a 3d SCFT with no continuous global symmetry. There are also two new non-supersymmetric solutions for which all 70 scalar fields in the supergravity theory have masses above the BF bound. All of these AdS$_4$ solutions can be uplifted to massive type IIA supergravity. Motivated by this we present the low lying operator spectra of the dual 3d CFTs for all known supersymmetric AdS$_4$ solutions in the theory and organize them into superconformal multiplets.

hep-th↗

arXiv2020

Temporal Coding in Spiking Neural Networks with Alpha Synaptic Function: Learning with Backpropagation

The timing of individual neuronal spikes is essential for biological brains to make fast responses to sensory stimuli. However, conventional artificial neural networks lack the intrinsic temporal coding ability present in biological networks. We propose a spiking neural network model that encodes information in the relative timing of individual neuron spikes. In classification tasks, the output of the network is indicated by the first neuron to spike in the output layer. This temporal coding scheme allows the supervised training of the network with backpropagation, using locally exact derivatives of the postsynaptic spike times with respect to presynaptic spike times. The network operates using a biologically-plausible alpha synaptic transfer function. Additionally, we use trainable synchronisation pulses that provide bias, add flexibility during training and exploit the decay part of the alpha function. We show that such networks can be trained successfully on noisy Boolean logic tasks and on the MNIST dataset encoded in time. The results show that the spiking neural network outperforms comparable spiking models on MNIST and achieves similar quality to fully connected conventional networks with the same architecture. We also find that the spiking network spontaneously discovers two operating regimes, mirroring the accuracy-speed trade-off observed in human decision-making: a slow regime, where a decision is taken after all hidden neurons have spiked and the accuracy is very high, and a fast regime, where a decision is taken very fast but the accuracy is lower. These results demonstrate the computational power of spiking networks with biological characteristics that encode information in the timing of individual neurons. By studying temporal coding in spiking networks, we aim to create building blocks towards energy-efficient and more complex biologically-inspired neural architectures.

cs.NE↗

arXiv2020

Intelligent Matrix Exponentiation

We present a novel machine learning architecture that uses the exponential of a single input-dependent matrix as its only nonlinearity. The mathematical simplicity of this architecture allows a detailed analysis of its behaviour, providing robustness guarantees via Lipschitz bounds. Despite its simplicity, a single matrix exponential layer already provides universal approximation properties and can learn fundamental functions of the input, such as periodic functions or multivariate polynomials. This architecture outperforms other general-purpose architectures on benchmark problems, including CIFAR-10, using substantially fewer parameters.

cs.LG↗

arXiv2020

A Cornucopia of AdS$_5$ Vacua

We report on a systematic search for AdS$_5$ vacua corresponding to critical points of the potential in the five-dimensional $\mathcal{N}=8$ SO(6) gauged supergravity. By employing Google's TensorFlow Machine Learning library, we find the total of 32 critical points including 5 previously known ones. All 27 new critical points are non-supersymmetric. We compute the mass spectra of scalar fluctuatons for all points and find that the non-supersymmetric AdS$_5$ vacua are perturbatively unstable. Many of the new critical points can be found analytically within consistent truncations of the $\mathcal{N}=8$ supergravity with respect to discrete subgroups of the S(O(6)$\times$ GL(2,$\mathbb{R}$)) symmetry of the potential. In particular, we discuss in detail a $\mathbb{Z}_2^3$-invariant truncation with 10 scalar fields and 15 critical points. We also compute explicitly the scalar potential in a $\mathbb{Z}_2^2$-invariant extension of that truncation to 18 scalar fields and reproduce 17 of the 32 critical points from the numerical search. Finally, we show that the full potential as a function of 42 scalar fields can be studied analytically using the so-called solvable parametrization. In particular, we find that all critical points lie in a $\mathbb{Z}_2$-invariant subspace spanned by 22 scalar fields.

hep-th↗

arXiv2019

A new N=1 AdS4 Vacuum of Maximal Supergravity

The recent comprehensive numerical study of critical points of the scalar potential of four-dimensional N=8, SO(8) gauged supergravity using Machine Learning software has led to a discovery of a new N=1 vacuum with a triality-invariant SO(3) symmetry. Guided by the numerical data for that point, we obtain a consistent SO(3)xZ2-invariant truncation of the N=8 theory to an N=1 supergravity with three chiral multiplets. Critical points of the truncated scalar potential include both the N=1 point as well as two new non-supersymmetric and perturbatively unstable points not found by previous searches. Studying the structure of the submanifold of SO(3)xZ2-invariant supergravity scalars, we find that it has a simple interpretation as a submanifold of the 14-dimensional Z2^3-invariant scalar manifold (SU(1,1)/U(1))^7, for which we find a rather remarkable superpotential whose structure matches the single bit error correcting (7, 4) Hamming code. This 14-dimensional scalar manifold contains approximately one quarter of the known critical points. We also show that there exists a smooth supersymmetric domain wall which interpolates between the new N=1 AdS4 solution and the maximally supersymmetric AdS4 vacuum. Using holography, this result indicates the existence of an N=1 RG flow from the ABJM SCFT to a new strongly interacting conformal fixed point in the IR.

hep-th↗

arXiv2019

Committee Draft of JPEG XL Image Coding System

JPEG XL is a practical approach focused on scalable web distribution and efficient compression of high-quality images. It provides various benefits compared to existing image formats: 60% size reduction at equivalent subjective quality; fast, parallelizable decoding and encoding configurations; features such as progressive, lossless, animation, and reversible transcoding of existing JPEG with 22% size reduction; support for high-quality applications including wide gamut, higher resolution/bit depth/dynamic range, and visually lossless coding. The JPEG XL architecture is traditional block-transform coding with upgrades to each component.

eess.IV↗

arXiv2019

SO(8) Supergravity and the Magic of Machine Learning

Using de Wit-Nicolai $D=4\;\mathcal{N}=8\;SO(8)$ supergravity as an example, we show how modern Machine Learning software libraries such as Google's TensorFlow can be employed to greatly simplify the analysis of high-dimensional scalar sectors of some M-Theory compactifications. We provide detailed information on the location, symmetries, and particle spectra and charges of 192 critical points on the scalar manifold of SO(8) supergravity, including one newly discovered $\mathcal{N}=1$ vacuum with $SO(3)$ residual symmetry, one new potentially stabilizable non-supersymmetric solution, and examples for "Galois conjugate pairs" of solutions, i.e. solution-pairs that share the same gauge group embedding into~$SO(8)$ and minimal polynomials for the cosmological constant. Where feasible, we give analytic expressions for solution coordinates and cosmological constants. As the authors' aspiration is to present the discussion in a form that is accessible to both the Machine Learning and String Theory communities and allows adopting our methods towards the study of other models, we provide an introductory overview over the relevant Physics as well as Machine Learning concepts. This includes short pedagogical code examples. In particular, we show how to formulate a requirement for residual Supersymmetry as a Machine Learning loss function and effectively guide the numerical search towards supersymmetric critical points. Numerical investigations suggest that there are no further supersymmetric vacua beyond this newly discovered fifth solution.

hep-th↗

arXiv2012

FlowPy - a numerical solver for functional renormalization group equations

FlowPy is a numerical toolbox for the solution of partial differential equations encountered in Functional Renormalization Group equations. This toolbox compiles flow equations to fast machine code and is able to handle coupled systems of flow equations with full momentum dependence, which furthermore may be given implicitly.

physics.comp-ph↗

arXiv2011

Casimir Forces via Worldline Numerics: Method Improvements and Potential Engineering Applications

The string theory inspired Worldline Numerics approach to Casimir force calculations has some favourable characteristics that might make it well suited for geometric optimization problems as they arise e.g. in NEMS device engineering. We explain this aspect in detail, developing some refinements of the method along the way. Also, we comment on the problem of generalizing Worldline Numerics from scalars to photons in the presence of conductors.

hep-th↗

arXiv2011

The Encyclopedic Reference of Critical Points for SO(8)-Gauged N=8 Supergravity Part 1: Cosmological Constants in the Range -Λ/g^2 \in [6;14.7)

This article is part of a collection that strives to collect and provide in an unified form data about all the critical points on the scalar manifold of SO(8)-gauged N=8 supergravity in four dimensions known so far. The vast majority of these were obtained using the enhanced sensitivity backpropagation method introduced by the author in 2008. This part of the collection describes 41 critical points, 7 of which have been known for more than two decades, 8 of which were discovered recently, and 26 are novel. The residual gauge symmetries of these 41 critical points (likely) are SO(8) with N=8 SUSY (1x), SO(7) (2x), SU(4) (1x), G2 with N=1 SUSY (1x), SU(3)xU(1) with N=2 SUSY (1x), SO(3)xSO(3) (2x), SO(3)xU(1)xU(1) (1x), SO(3)xU(1) (3x), SO(3) (3x), U(1)xU(1) with N=1 SUSY (1x), U(1)xU(1) without SUSY (4x), U(1) (11x), and None (10x). Analytic conjectures (not yet proven but overwhelmingly likely correct) are given for the locations and cosmological constants of some critical points.

hep-th↗