arXiv subjects

Jannes Gladrow

Publications and source records attributed to Jannes Gladrow.

10 recordsLinked to original sources

arXiv2026

Amortized Inference of Causal Models via Conditional Fixed-Point Iterations

Structural Causal Models (SCMs) offer a principled framework to reason about interventions and support out-of-distribution generalization, which are key goals in scientific discovery. However, the task of learning SCMs from observed data poses formidable challenges, and often requires training a separate model for each dataset. In this work, we propose an amortized inference framework that trains a single model to predict the causal mechanisms of SCMs conditioned on their observational data and causal graph. We first use a transformer-based architecture for amortized learning of dataset embeddings, and then extend the Fixed-Point Approach (FiP) to infer the causal mechanisms conditionally on their dataset embeddings. As a byproduct, our method can generate observational and interventional data from novel SCMs at inference time, without updating parameters. Empirical results show that our amortized procedure performs on par with baselines trained specifically for each dataset on both in and out-of-distribution problems, and also outperforms them in scarce data regimes.

cs.LG↗

arXiv2025

Implicit Language Models are RNNs: Balancing Parallelization and Expressivity

State-space models (SSMs) and transformers dominate the language modeling landscape. However, they are constrained to a lower computational complexity than classical recurrent neural networks (RNNs), limiting their expressivity. In contrast, RNNs lack parallelization during training, raising fundamental questions about the trade off between parallelization and expressivity. We propose implicit SSMs, which iterate a transformation until convergence to a fixed point. Theoretically, we show that implicit SSMs implement the non-linear state-transitions of RNNs. Empirically, we find that only approximate fixed-point convergence suffices, enabling the design of a scalable training curriculum that largely retains parallelization, with full convergence required only for a small subset of tokens. Our approach demonstrates superior state-tracking capabilities on regular languages, surpassing transformers and SSMs. We further scale implicit SSMs to natural language reasoning tasks and pretraining of large-scale language models up to 1.3B parameters on 207B tokens representing, to our knowledge, the largest implicit model trained to date. Notably, our implicit models outperform their explicit counterparts on standard benchmarks. Our code is publicly available at http://github.com/microsoft/implicit_languagemodels .

cs.LG↗

arXiv2024

Resolution dependence of most probable pathways with state-dependent diffusivity

Recent experiments have probed the relative likelihoods of trajectories in stochastic systems by observing survival probabilities within a tube of radius $R$ in spacetime. We measure such probabilities here for a colloidal particle in a corrugated channel, corresponding to a bistable potential with state-dependent diffusivity. In contrast to previous findings for state-independent noise, we find that the most probable pathway changes qualitatively as the tube radius $R$ is altered. We explain this by computing the survival probabilities predicted by overdamped Langevin dynamics. At high enough resolution (small enough $R$), survival probabilities depend solely on diffusivity variations, independent of deterministic forces; finite $R$ corrections yield a generalization of the Onsager-Machlup action. As corollary, ratios of survival probabilities are singular as $R \to 0$, but become regular, and described by the classical Onsager-Machlup action, only in the special case of state-independent noise.

cond-mat.stat-mech↗

arXiv2023

Efficient data transport over multimode light-pipes with Megapixel images using differentiable ray tracing and Machine-learning

Retrieving images transmitted through multi-mode fibers is of growing interest, thanks to their ability to confine and transport light efficiently in a compact system. Here, we demonstrate machine-learning-based decoding of large-scale digital images (pages), maximizing page capacity for optical storage applications. Using a millimeter-sized square cross-section waveguide, we image an 8-bit spatial light modulator, presenting data as a matrix of symbols. Normally, decoders will incur a prohibitive O(n^2) computational scaling to decode n symbols in spatially scrambled data. However, by combining a digital twin of the setup with a U-Net, we can retrieve up to 66 kB using efficient convolutional operations only. We compare trainable ray-tracing-based with eigenmode-based twins and show the former to be superior thanks to its ability to overcome the simulation-to-experiment gap by adjusting to optical imperfections. We train the pipeline end-to-end using a differentiable mutual-information estimator based on the von-Mises distribution, generally applicable to phase-coding channels.

physics.optics↗

arXiv2021

Experimental measurement of relative path probabilities and stochastic actions

For diffusive stochastic dynamics, the probability to observe any individual trajectory is vanishingly small, making it unclear how to experimentally validate theoretical results for ratios of path probabilities. We provide the missing link between theory and experiment, by establishing a protocol to extract ratios of path probabilities from measured time series. For experiments on a single colloidal particle in a microchannel, we extract both ratios of path probabilities, and the most probable path for a barrier crossing, and find excellent agreement with independently calculated predictions based on the Onsager-Machlup stochastic action. Our experimental results at room temperature are found to be inconsistent with the low-noise Freidlin-Wentzell stochastic action, and we discuss under which circumstances the latter is expected to describe the most probable path. Furthermore, while the experimentally accessible ratio of path probabilities is uniquely determined, the formal path-integral action is known to depend on the time-discretization scheme used for deriving it; we reconcile these two seemingly contradictory facts by careful analysis of the time-slicing derivation of the path integral. Our experimental protocol enables us to probe probability distributions on path space, and allows us to relate theoretical single-trajectory results to measurement.

cond-mat.stat-mech↗

arXiv2020

Experimental evidence of symmetry breaking of transition-path times

While Kramers' rates have been studied for almost a century, the transition path time between states has only recently received attention. Transition paths between different energy levels are expected to be indistinguishable in shape and have equal uphill and downhill lengths. This fundamental symmetry often prevents directional sensing in experiments. Here, we report experimental evidence for transition path time symmetry and its breakdown on the single-molecule and mesoscale. In automated Brownian dynamics experiments, we establish first-passage time symmetries of colloids driven by femtoNewton-range forces in holographically-created optical landscapes confined in microchannels. Conversely, we show that transitions which couple in a path-dependent manner to fluctuating forces exhibit asymmetry. We reproduce this asymmetry in folding transitions of DNA-hairpins driven out of equilibrium and suggest a topological mechanism for the symmetry breakdown. Our results can reveal directionality in molecular transitions or translocations through membrane channels and nanopores.

cond-mat.stat-mech↗

arXiv2019

Optimizing Brownian escape rates by potential shaping

Brownian escape is key to a wealth of physico-chemical processes, including polymer folding, and information storage. The frequency of thermally activated energy barrier crossings is assumed to generally decrease exponentially with increasing barrier height. Here, we show experimentally that higher, fine-tuned barrier profiles result in significantly enhanced escape rates in breach of the intuition relying on the above scaling law, and address in theory the corresponding conditions for maximum speed-up. Importantly, our barriers end on the same energy on which they start. For overdamped dynamics, the achievable boost of escape rates is, in principle, unbounded so that the barrier optimization has to be regularized. We derive optimal profiles under two different regularizations, and uncover the efficiency of N-shaped barriers. We then demonstrate the viability of such a potential in automated microfluidic Brownian dynamics experiments using holographic optical tweezers and achieve a doubling of escape rates compared to unhindered Brownian motion. Finally, we show that this escape rate boost extends into the low-friction inertial regime.

cond-mat.soft↗

arXiv2019

Digital phase-only holography using deep conditional generative models

Holographic wave-shaping has found numerous applications across the physical sciences, especially since the development of digital spatial-light modulators (SLMs). A key challenge in digital holography consists in finding optimal hologram patterns which transform the incoming laser beam into desired shapes in a conjugate optical plane. The existing repertoire of approaches to solve this inverse problem is built on iterative phase-retrieval algorithms, which do not take optical aberrations and deviations from theoretical models into account. Here, we adopt a physics-free, data-driven, and probabilistic approach to the problem. Using deep conditional generative models such as Generative-Adversarial Networks (cGAN) or Variational Autoencoder (cVAE), we approximate conditional distributions of holograms for a given target laser intensity pattern. In order to reduce the cardinality of the problem, we train our models on a proxy mapping relating an 8x8-matrix of complex-valued spatial-frequency coefficients to the ensuing 100x100-shaped intensity distribution recorded on a camera. We discuss the degree of 'ill-posedness' that remains in this reduced problem and compare different generative model architectures in terms of their ability to find holograms that reconstruct given intensity patterns. Finally, we challenge our models to generalise to synthetic target intensities, where the existence of matching holograms cannot be guaranteed. We devise a forward-interpolating training scheme aimed at providing models the ability to interpolate in laser intensity space, rather than hologram space and show that this indeed enhances model performance on synthetic data sets.

eess.IV↗

arXiv2017

Particle transport across a channel via an oscillating potential

Membrane protein transporters alternate their substrate-binding sites between the extracellular and cytosolic side of the membrane according to the alternating access mechanism. Inspired by this intriguing mechanism devised by nature, we study particle transport through a channel coupled with an energy well that oscillates its position between the two entrances of the channel. We optimize particle transport across the channel by adjusting the oscillation frequency. At the optimal oscillation frequency, the translocation rate through the channel is a hundred times higher with respect to free diffusion across the channel. Our findings reveal the effect of time dependent potentials on particle transport across a channel and will be relevant for membrane transport and microfluidics application.

physics.bio-ph↗

arXiv2017

Nonequilibrium dynamics of probe filaments in actin-myosin networks

Active dynamic processes of cells are largely driven by the cytoskeleton, a complex and adaptable semiflexible polymer network, motorized by mechanoenzymes. Small dimensions, confined geome- tries and hierarchical structures make it challenging to probe dynamics and mechanical response of such networks. Embedded semiflexible probe polymers can serve as non-perturbing multi-scale probes to detect force distributions in active polymer networks. We show here that motor-induced forces transmitted to the probe polymers are reflected in non-equilibrium bending dynamics, which we analyze in terms of spatial eigenmodes of an elastic beam. We demonstrate how these active forces induce correlations among these mode amplitudes, which furthermore break time-reversal symmetry. This leads to a breaking of detailed balance in this mode space. We derive analytical predictions for the magnitude of resulting probability currents in mode space in the white-noise limit of motor activity. We relate the structure of these currents to the spatial profile of motor- induced forces along the probe polymers and provide a general relation for observable currents on two-dimensional hyperplanes.

physics.bio-ph↗