arXiv subjects

Nicolas Deutschmann

Publications and source records attributed to Nicolas Deutschmann.

16 recordsLinked to original sources

arXiv2026

EvoFlows: Evolutionary Edit-Based Flow-Matching for Protein Engineering

We introduce EvoFlows, a variable-length protein sequence-to-sequence modeling approach designed for protein engineering. Existing protein language models are poorly suited for optimization tasks: autoregressive models require full sequence generation, masked language and discrete diffusion models rely on pre-specified mutation locations, and no existing methods naturally support insertions and deletions relative to a template sequence. EvoFlows learns mutational trajectories between evolutionarily related protein sequences via edit flows, allowing it to perform a controllable number of mutations (insertions, deletions, and substitutions) on a template sequence, predicting not only _which_ mutation to perform, but also _where_ it should occur. Through extensive _in silico_ evaluation on diverse protein families from UniRef and OAS, we show that EvoFlows generates variants that remain consistent with natural protein families while exploring farther from template sequences than leading baselines.

cs.LG↗

arXiv2025

g-DPO: Scalable Preference Optimization for Protein Language Models

Direct Preference Optimization (DPO) is an effective approach for aligning protein language models with experimental design goals. However, DPO faces a scalability bottleneck: the number of possible training pairs grows quadratically with the number of labeled sequences, leading to prohibitive training times even for modestly sized datasets. We introduce g-DPO, a framework that (i) uses sequence space clustering to prune redundant pairs while preserving training signal, and (ii) amortizes likelihood computations with group-based approximations. Across three protein engineering tasks, g-DPO maintains in silico and in vitro performance that is statistically indistinguishable from standard DPO, while converging 1.7x to 5.4x times faster, with speedups that scale with dataset size and the structure of the underlying mutational landscape.

cs.LG↗

arXiv2024

Accelerating HEP simulations with Neural Importance Sampling

Many high-energy-physics (HEP) simulations for the LHC rely on Monte Carlo using importance sampling by means of the VEGAS algorithm. However, complex high-precision calculations have become a challenge for the standard toolbox, as this approach suffers from poor performance in complex cases. As a result, there has been keen interest in HEP for modern machine learning to power adaptive sampling. While previous studies have shown the potential of normalizing-flow-powered neural importance sampling (NIS) over VEGAS, there remains a gap in accessible tools tailored for non-experts. In response, we introduce ZüNIS, a fully automated NIS library designed to bridge this divide, while at the same time providing the infrastructure to customise the algorithm for dealing with challenging tasks. After a general introduction on NIS, we first show how to extend the original formulation of NIS to reuse samples over multiple gradient steps while guaranteeing a stable training, yielding a significant improvement for slow functions. Next, we introduce the structure of the library, which can be used by non-experts with minimal effort and is extensivly documented, which is crucial to become a mature tool for the wider HEP public. We present systematic benchmark results on both toy and physics examples, and stress the benefit of providing different survey strategies, which allows higher performance in challenging cases. We show that ZüNIS shows high performance on a range of problems with limited fine-tuning.

hep-ph↗

arXiv2023

Conformal Autoregressive Generation: Beam Search with Coverage Guarantees

We introduce two new extensions to the beam search algorithm based on conformal predictions (CP) to produce sets of sequences with theoretical coverage guarantees. The first method is very simple and proposes dynamically-sized subsets of beam search results but, unlike typical CP procedures, has an upper bound on the achievable guarantee depending on a post-hoc calibration measure. Our second algorithm introduces the conformal set prediction procedure as part of the decoding process, producing a variable beam width which adapts to the current uncertainty. While more complex, this procedure can achieve coverage guarantees selected a priori. We provide marginal coverage bounds for each method, and evaluate them empirically on a selection of tasks drawing from natural language processing and chemistry.

cs.LG↗

arXiv2023

Adaptive Conformal Regression with Jackknife+ Rescaled Scores

Conformal regression provides prediction intervals with global coverage guarantees, but often fails to capture local error distributions, leading to non-homogeneous coverage. We address this with a new adaptive method based on rescaling conformal scores with an estimate of local score distribution, inspired by the Jackknife+ method, which enables the use of calibration data in conformal scores without breaking calibration-test exchangeability. Our approach ensures formal global coverage guarantees and is supported by new theoretical results on local coverage, including an a posteriori bound on any calibration score. The strength of our approach lies in achieving local coverage without sacrificing calibration set size, improving the applicability of conformal prediction intervals in various settings. As a result, our method provides prediction intervals that outperform previous methods, particularly in the low-data regime, making it especially relevant for real-world applications such as healthcare and biomedical domains where uncertainty needs to be quantified accurately despite low sample data.

cs.LG↗

arXiv2022

Attention-based Interpretable Regression of Gene Expression in Histology

Interpretability of deep learning is widely used to evaluate the reliability of medical imaging models and reduce the risks of inaccurate patient recommendations. For models exceeding human performance, e.g. predicting RNA structure from microscopy images, interpretable modelling can be further used to uncover highly non-trivial patterns which are otherwise imperceptible to the human eye. We show that interpretability can reveal connections between the microscopic appearance of cancer tissue and its gene expression profiling. While exhaustive profiling of all genes from the histology images is still challenging, we estimate the expression values of a well-known subset of genes that is indicative of cancer molecular subtype, survival, and treatment response in colorectal cancer. Our approach successfully identifies meaningful information from the image slides, highlighting hotspots of high gene expression. Our method can help characterise how gene expression shapes tissue morphology and this may be beneficial for patient stratification in the pathology unit. The code is available on GitHub.

q-bio.QM↗

arXiv2022

Is Attention Interpretation? A Quantitative Assessment On Sets

The debate around the interpretability of attention mechanisms is centered on whether attention scores can be used as a proxy for the relative amounts of signal carried by sub-components of data. We propose to study the interpretability of attention in the context of set machine learning, where each data point is composed of an unordered collection of instances with a global label. For classical multiple-instance-learning problems and simple extensions, there is a well-defined "importance" ground truth that can be leveraged to cast interpretation as a binary classification problem, which we can quantitatively evaluate. By building synthetic datasets over several data modalities, we perform a systematic assessment of attention-based interpretations. We find that attention distributions are indeed often reflective of the relative importance of individual instances, but that silent failures happen where a model will have high classification performance but attention patterns that do not align with expectations. Based on these observations, we propose to use ensembling to minimize the risk of misleading attention-based explanations.

cs.LG↗

arXiv2020

Top-Yukawa contributions to bbH production at the LHC

We study the production of a Higgs boson in association with bottom quarks ($b\bar{b}H$) in hadronic collisions at the LHC, including the different contributions stemming from terms proportional to the top-quark Yukawa coupling ($y_t^2$), to the bottom-quark one ($y_b^2$), and to their interference ($y_b y_t$). Our results are accurate to next-to-leading order in QCD, employ the four-flavour scheme and the (Born-improved) heavy-top quark approximation. We find that next-to-leading order corrections to the $y_t^2$ component are sizable, making it the dominant production mechanism for associated $b\bar{b}H$ production in the Standard Model and increasing its inclusive rate by almost a factor of two. By studying final-state distributions of the various contributions, we identify observables and selection cuts that can be used to select the various components and to improve the experimental sensitivity of $b\bar{b}H$ production on the bottom-quark Yukawa coupling.

hep-ph↗

arXiv2020

Quark mass effects in two-loop Higgs amplitudes

We provide two two-loop amplitudes relevant for precision Higgs physics. The first is the two-loop amplitude for Higgs boson production through gluon fusion with exact dependence on the top quark mass up to squared order in the dimensional regulator $ε$. The second result we provide is the two-loop amplitude for the decay of a Higgs boson into a pair of massive bottom quarks through the Higgs-to-gluon coupling in the infinite top mass limit. Both amplitudes are computed by finding canonical bases of master integrals, which we evaluate explicitly in terms of harmonic polylogarithms. We obtain the bare, renormalized and IR-subtracted amplitude and provide the results in terms of building blocks suitable to changing renormalization schemes.

hep-ph↗

arXiv2019

Momentum mappings for subtractions at higher orders in QCD

Subtraction schemes provide a systematic way to compute fully-differential cross sections beyond the leading order in the strong coupling constant. These methods make singular real-emission corrections integrable in phase space by the addition of suitable counterterms. Such counterterms may be defined using momentum mappings, which are parametrisations of the phase space that factorise the variables that describe the particles becoming unresolved in some infrared or collinear limit from the variables that describe an on-shell phase space for the resolved particles. In this work, we review existing momentum mappings in a unified framework and introduce new ones for final-collinear and soft counterterms. The new mappings work in the presence of massive particles and with an arbitrary number of soft particles or of clusters of collinear particles, making them fit for subtraction methods at any order in perturbation theory. The new mapping for final-collinear counterterms is also used to elucidate relations among existing final-collinear mappings.

hep-ph↗

arXiv2018

Gluon-fusion Higgs production in the Standard Model Effective Field Theory

We provide the complete set of predictions needed to achieve NLO accuracy in the Standard Model Effective Field Theory at dimension six for Higgs production in gluon fusion. In particular, we compute for the first time the contribution of the chromomagnetic operator $ \bar Q_L Φσq_R G$ at NLO in QCD, which entails two-loop virtual and one-loop real contributions, as well as renormalisation and mixing with the Yukawa operator $Φ^\dagger Φ\, \bar Q_L Φq_R$ and the gluon-fusion operator $Φ^\dagger Φ\, GG$. Focusing on the top-quark-Higgs couplings, we consider the phenomenological impact of the NLO corrections in constraining the three relevant operators by implementing the results into the MadGraph5_aMC@NLO framework. This allows us to compute total cross sections as well as to perform event generation at NLO that can be directly employed in experimental analyses.

hep-ph↗

arXiv2017

Current LHC Constraints on Minimal Universal Extra Dimensions

In this letter, we present LHC limits on the minimal universal extra dimension (MUED) model from LHC Run 1 data and current limits from searches of the ongoing Run 2. Typical collider signals of the Kaluza-Klein (KK) states mimic generic degenerate supersymmetry (SUSY) missing transverse momentum signatures since the excited KK particles cascade decay to jets, leptons and the lightest KK particle which is stable due to KK parity and thus evades detection. We test the parameter space against a large number of supersymmetry based missing energy searches implemented in the public code CheckMATE. We demonstrate the complementarity of employing various searches which target a large number of final state signatures, and we derive the most up to date limits on the MUED parameter space from 13 TeV SUSY searches.

hep-ph↗

arXiv2017

Compact Extra Dimensions in Quantum Mechanics

Extra-dimensions are a common topic in popular descriptions of theoretical physics with which undergraduate student most often have no contact in physics courses. This paper shows how students could be introduced to this topic by presenting an approach to two basic consequences of the presence of compact extra-dimensions based on undergraduate-level physics. The insensibility of low-energy physics to compact extra dimensions is illustrated in the context of non-relativistic quantum mechanics and the prediction of Kaluza-Klein excitations of particles is discussed in the framework of relativistic wave-equations. An exercise that could be used as a follow-up to the "particle in a box" is proposed.

physics.gen-ph↗

arXiv2016

Towards Kaluza-Klein Dark Matter on Nilmanifolds

We present a first study of the field spectrum on a class of negatively-curved compact spaces: nilmanifolds or twisted tori. This is a case where analytical results can be obtained, allowing to check numerical methods. We focus on the Kaluza-Klein expansion of a scalar field. The results are then applied to a toy model where a natural Dark Matter candidate arises as a stable massive state of the bulk scalar.

hep-th↗

arXiv2016

Dark matter and localised fermions from spherical orbifolds?

We study a class of six-dimensional models based on positive curvature surfaces (spherical 2-orbifolds) as extra-spaces. Using the Newman-Penrose formalism, we discuss the particle spectrum in this class of models. The fermion spectrum problem, which has been addressed with flux compactifications in the past, can be avoided using localised fermions. In this framework, we find that there are four types of geometry compatible with the existence of a stable dark matter candidate and we study the simplest case in detail. Using the complementarity between collider resonance searches and relic density constraints, we show that this class of models is under tension, unless the model lies in a funnel region characterised by a resonant Higgs s-channel in the dark matter annihilation.

hep-ph↗

arXiv2014

Multi-tops at the LHC

The experiments at the LHC are searching for many different final states that can hint to the presence of new physics beyond the Standard Model. One of the most interesting and promising sectors for these searches is that of the top quark, for both theoretical and phenomenological reasons linked to its large mass and to its possible special role in the electroweak symmetry breaking sector. We suggest that multi-top events, beyond the standard $t$-$\bar t$ and four top searches, can bring further insight in constraining and discovering physics beyond the Standard Model, taking advantage of experimental techniques similar to those used in present top-quark analyses. This is relevant both for the next data taking runs at the LHC and even more at higher luminosity and higher energy collider options, which are discussed for future LHC upgrades and future accelerators. In particular we consider six top and eight top final states, discussing the generic colour representations for beyond the Standard Model particles giving rise to those final state. We also discuss the limits which can be extracted by using the present analyses sensitive to four top final states, as well as the potential bounds from new searches we propose to experimental collaborations as an alternative.

hep-ph↗