arXiv subjects

Sam Dillavou

Publications and source records attributed to Sam Dillavou.

At least 19 recordsLinked to original sources

arXiv2025

Physical Reinforcement Learning

Digital computers are power-hungry and largely intolerant of damaged components, making them potentially difficult tools for energy-limited autonomous agents in uncertain environments. Recently developed Contrastive Local Learning Networks (CLLNs) - analog networks of self-adjusting nonlinear resistors - are inherently low-power and robust to physical damage, but were constructed to perform supervised learning. In this work we demonstrate success on two simple RL problems using Q-learning adapted for simulated CLLNs. Doing so makes explicit the components (beyond the network being trained) required to enact various tools in the RL toolbox, some of which (policy function and value function) are more natural in this system than others (replay buffer). We discuss assumptions such as the physical safety that digital hardware requires, CLLNs can forgo, and biological systems cannot rely on, and highlight secondary goals that are important in biology and trainable in CLLNs, but make little sense in digital computers.

cs.LG↗

arXiv2025

Analog Physical Systems Can Exhibit Double Descent

An important component of the success of large AI models is double descent, in which networks avoid overfitting as they grow relative to the amount of training data, instead improving their performance on unseen data. Here we demonstrate double descent in a decentralized analog network of self-adjusting resistive elements. This system trains itself and performs tasks without a digital processor, offering potential gains in energy efficiency and speed -- but must endure component non-idealities. We find that standard training fails to yield double descent, but a modified protocol that accommodates this inherent imperfection succeeds. Our findings show that analog physical systems, if appropriately trained, can exhibit behaviors underlying the success of digital AI. Further, they suggest that biological systems might similarly benefit from over-parameterization.

cond-mat.dis-nn↗

arXiv2025

Understanding and Embracing Imperfection in Physical Learning Networks

Performing machine learning with analog signals offers advantages in speed and energy efficiency, but sensitivity to component and measurement imperfections often foils training without a system-specific companion digital model. Here we take a different perspective, accepting and characterizing these inherent imperfections and ultimately overcoming them without digital models. We train an analog network of self-adjusting resistors -- a contrastive local learning network -- for multiple tasks, and observe limit cycles and scaling behaviors that limit precision, erase memory of previous tasks, and are absent in `perfect' systems. We develop an analytical model capturing these phenomena as a consequence of an uncontrolled learning bias continuously modifying the underlying representation of learned tasks, reminiscent of representational drift in the brain. Finally, we introduce and demonstrate a system-agnostic training method that greatly suppresses these effects. Our work points to a new, scalable analog approach that eschews precise modeling and instead thrives in the mess of real systems.

cond-mat.dis-nn↗

arXiv2024

Cornerstones are the Key Stones: Using Interpretable Machine Learning to Probe the Clogging Process in 2D Granular Hoppers

The sudden arrest of flow by formation of a stable arch over an outlet is a unique and characteristic feature of granular materials. Previous work suggests that grains near the outlet randomly sample configurational flow microstates until a clog-causing flow microstate is reached. However, factors that lead to clogging remain elusive. Here we experimentally observe over 50,000 clogging events for a tridisperse mixture of quasi-2D circular grains, and utilize a variety of machine learning (ML) methods to search for predictive signatures of clogging microstates. This approach fares just modestly better than chance. Nevertheless, our analysis using linear Support Vector Machines (SVMs) highlights the position of potential arch cornerstones as a key factor in clogging likelihood. We verify this experimentally by varying the position of a fixed (cornerstone) grain, and show that such a grain dictates the size of feasible flow-ending arches, and thus the time and mass of each flow. Positioning this grain correctly can even increase the ejected mass by over 50%. Our findings demonstrate that interpretable ML algorithms like SVMs can uncover meaningful physics even when their predictive power is below the standards of conventional ML practice.

cond-mat.soft↗

arXiv2024

Comparing the information content of probabilistic representation spaces

Probabilistic representation spaces convey information about a dataset and are shaped by factors such as the training data, network architecture, and loss function. Comparing the information content of such spaces is crucial for understanding the learning process, yet most existing methods assume point-based representations, neglecting the distributional nature of probabilistic spaces. To address this gap, we propose two information-theoretic measures to compare general probabilistic representation spaces by extending classic methods to compare the information content of hard clustering assignments. Additionally, we introduce a lightweight method of estimation that is based on fingerprinting a representation space with a sample of the dataset, designed for scenarios where the communicated information is limited to a few bits. We demonstrate the utility of these measures in three case studies. First, in the context of unsupervised disentanglement, we identify recurring information fragments within individual latent dimensions of VAE and InfoGAN ensembles. Second, we compare the full latent spaces of models and reveal consistent information content across datasets and methods, despite variability during training. Finally, we leverage the differentiability of our measures to perform model fusion, synthesizing the information content of weak learners into a single, coherent representation. Across these applications, the direct comparison of information content offers a natural basis for characterizing the processing of information.

cs.LG↗

arXiv2024

The Equation of Motion for Taut-Line Buzzers

Equations of motion are developed for the oscillatory rotation of a disk suspended between twisted strings kept under tension by a hanging mass, to which additional forces may be applied. In the absence of forcing, damped harmonic oscillations are observed to decay with an exponential time envelope for two different string types. This is consistent with damping caused by string viscosity, rather than air turbulence, and may be quantified in terms of a quality factor. To test the proposed equation of motion and model for viscous damping within the string, we measure both the natural oscillation frequency and the quality factor for widely varied values of string length, string radius, disk moment of inertia, and hanging mass. The data are found to scale in good accord with predictions. A variation where rotational kinetic energy is converted back and forth to spring potential energy is also discussed.

cond-mat.soft↗

arXiv2023

Stochastic dynamics of granular hopper flows: a configurational mode controls the stability of clogs

Granular flows in small-outlet hoppers exhibit several characteristic but poorly understood behaviors: temporary clogs (pauses) where flow stops before later spontaneously restarting, permanent clogs that last indefinitely, and non-Gaussian, non-monotonic flow-rate statistics. These aspects have been studied independently, but a model of hopper flow that includes all three has not been formulated. Here, we introduce a phenomenological model that provides a unifying dynamical explanation of all three behaviors: coupling between the flow rate and a hidden mode that controls the stability of clogs. In the theory, flow rate evolves according to Langevin dynamics with multiplicative noise and an absorbing state at zero flow, conditional on the hidden mode. The model fully reproduces the statistics of pause and clog events of a large ($>40,000$ flows) experimental dataset, including non-exponentially distributed clogging times and non-Gaussian flow rate distribution, and explains the stretched-exponential growth of the average clogging time with outlet size. Further, we identify the physical nature of the hidden mode in microscopic configurational features, including size and smoothness of the static arch structure formed during pauses and clogs. Our work provides a unifying framework for several poorly understood clogging phenomena, and suggests numerous new paths toward further understanding of this complex system.

cond-mat.soft↗

arXiv2023

Machine Learning Without a Processor: Emergent Learning in a Nonlinear Electronic Metamaterial

Standard deep learning algorithms require differentiating large nonlinear networks, a process that is slow and power-hungry. Electronic learning metamaterials offer potentially fast, efficient, and fault-tolerant hardware for analog machine learning, but existing implementations are linear, severely limiting their capabilities. These systems differ significantly from artificial neural networks as well as the brain, so the feasibility and utility of incorporating nonlinear elements have not been explored. Here we introduce a nonlinear learning metamaterial -- an analog electronic network made of self-adjusting nonlinear resistive elements based on transistors. We demonstrate that the system learns tasks unachievable in linear systems, including XOR and nonlinear regression, without a computer. We find our nonlinear learning metamaterial reduces modes of training error in order (mean, slope, curvature), similar to spectral bias in artificial neural networks. The circuitry is robust to damage, retrainable in seconds, and performs learned tasks in microseconds while dissipating only picojoules of energy across each transistor. This suggests enormous potential for fast, low-power computing in edge systems like sensors, robotic controllers, and medical devices, as well as manufacturability at scale for performing and studying emergent learning.

cond-mat.soft↗

arXiv2023

Physical learning of power-efficient solutions

As the size and ubiquity of artificial intelligence and computational machine learning (ML) models grow, their energy consumption for training and use is rapidly becoming economically and environmentally unsustainable. Neuromorphic computing, or the implementation of ML in hardware, has the potential to reduce this cost. In particular, recent laboratory prototypes of self-learning electronic circuits, examples of ``physical learning machines," open the door to analog hardware that directly employs physics to learn desired functions from examples. In this work, we show that this hardware platform allows for even further reduction of energy consumption by using good initial conditions as well as a new learning algorithm. Using analytical calculations, simulation and experiment, we show that a trade-off emerges when learning dynamics attempt to minimize both the error and the power consumption of the solution--greater power reductions can be achieved at the cost of decreasing solution accuracy. Finally, we demonstrate a practical procedure to weigh the relative importance of error and power minimization, improving power efficiency given a specific tolerance to error.

cond-mat.dis-nn↗

arXiv2023

Bellybutton: Accessible and Customizable Deep-Learning Image Segmentation

The conversion of raw images into quantifiable data can be a major hurdle in experimental research, and typically involves identifying region(s) of interest, a process known as segmentation. Machine learning tools for image segmentation are often specific to a set of tasks, such as tracking cells, or require substantial compute or coding knowledge to train and use. Here we introduce an easy-to-use (no coding required), image segmentation method, using a 15-layer convolutional neural network that can be trained on a laptop: Bellybutton. The algorithm trains on user-provided segmentation of example images, but, as we show, just one or even a portion of one training image can be sufficient in some cases. We detail the machine learning method and give three use cases where Bellybutton correctly segments images despite substantial lighting, shape, size, focus, and/or structure variation across the regions(s) of interest. Instructions for easy download and use, with further details and the datasets used in this paper are available at pypi.org/project/Bellybuttonseg.

cs.CV↗

arXiv2022

Aqueous foams in microgravity, measuring bubble sizes

The paper describes a study of wet foams in microgravity whose bubble size distribution evolves due to diffusive gas exchange. We focus on the comparison between the size of bubbles determined from images of the foam surface and the size of bubbles in the bulk foam, determined from Diffuse Transmission Spectroscopy (DTS). Extracting the bubble size distribution from images of a foam surface is difficult so we have used three different procedures : manual analysis, automatic analysis with a customized Python script and machine learning analysis. Once various pitfalls were identified and taken into account, all the three procedures yield identical results within error bars. DTS only allows the determination of an average bubble radius which is proportional to the photon transport mean free path $\ell^*$. The relation between the measured diffuse transmitted light intensity and {$\ell^*$} previously derived for slab-shaped samples of infinite lateral extent does not apply to the cuboid geometry of the cells used in the microgravity experiment. A new more general expression of the diffuse intensity transmitted with specific optical boundary conditions has been derived and applied to determine the average bubble radius. The temporal evolution of the average bubble radii deduced from DTS and of the same average radii of the bubbles measured at the sample surface are in very good agreement throughout the coarsening. Finally, ground experiments were performed to compare bubble size distributions in a bulk wet foam and at its surface at times so short that diffusive gas exchange is insignificant. They were found to be similar, confirming that bubbles seen at the surface are representative of the bulk foam bubbles.

cond-mat.soft↗

arXiv2022

Desynchronous Learning in a Physics-Driven Learning Network

In a neuron network, synapses update individually using local information, allowing for entirely decentralized learning. In contrast, elements in an artificial neural network (ANN) are typically updated simultaneously using a central processor. Here we investigate the feasibility and effect of desynchronous learning in a recently introduced decentralized, physics-driven learning network. We show that desynchronizing the learning process does not degrade performance for a variety of tasks in an idealized simulation. In experiment, desynchronization actually improves performance by allowing the system to better explore the discretized state space of solutions. We draw an analogy between desynchronization and mini-batching in stochastic gradient descent, and show that they have similar effects on the learning process. Desynchronizing the learning process establishes physics-driven learning networks as truly fully distributed learning machines, promoting better performance and scalability in deployment.

cond-mat.soft↗

arXiv2021

Physical learning beyond the quasistatic limit

Physical networks, such as biological neural networks, can learn desired functions without a central processor, using local learning rules in space and time to learn in a fully distributed manner. Learning approaches such as equilibrium propagation, directed aging, and coupled learning similarly exploit local rules to accomplish learning in physical networks such as mechanical, flow, or electrical networks. In contrast to certain natural neural networks, however, such approaches have so far been restricted to the quasistatic limit, where they learn on time scales slow compared to their physical relaxation. This quasistatic constraint slows down learning, limiting the use of these methods as machine learning algorithms, and potentially restricting physical networks that could be used as learning platforms. Here we explore learning in an electrical resistor network that implements coupled learning, both in the lab and on the computer, at rates that range from slow to far above the quasistatic limit. We find that up to a critical threshold in the ratio of the learning rate to the physical rate of relaxation, learning speeds up without much change of behavior or error. Beyond the critical threshold, the error exhibits oscillatory dynamics but the networks still learn successfully.

cond-mat.dis-nn↗

arXiv2021

Spatters and Spills: Spreading Dynamics for Partially Wetting Droplets

We present a solvable model inspired by dimensional analysis for the time-dependent spreading of droplets that partially wet a substrate, where the spreading eventually stops and the contact angle reaches a nonzero equilibrium value. We separately consider small droplets driven by capillarity and large droplets driven by gravity. To explore both regimes, we first measure the equilibrium radius versus a comprehensive range of droplet volumes for four household fluids, and we compare the results with predictions based on minimizing the sum of gravitational and interfacial energies. The agreement is good, and gives a reliable measurement of an equilibrium contact angle that is consistent in both small and large droplet regimes. Next we use energy considerations to develop equations of motion for the time dependence of the spreading, in both regimes, where the driving forces are balanced against viscous drag in the bulk of the droplet and by friction at the moving contact line. Our approach leads to explicit prediction of the functional form of the spreading dynamics. It successfully describes prior data for a small capillary-driven droplet, and it fits well to new data we obtain for large gravity-driven droplets with a wide range of volumes.

physics.flu-dyn↗

arXiv2021

Demonstration of Decentralized, Physics-Driven Learning

In typical artificial neural networks, neurons adjust according to global calculations of a central processor, but in the brain neurons and synapses self-adjust based on local information. Contrastive learning algorithms have recently been proposed to train physical systems, such as fluidic, mechanical, or electrical networks, to perform machine learning tasks from local evolution rules. However, to date such systems have only been implemented in silico due to the engineering challenge of creating elements that autonomously evolve based on their own response to two sets of global boundary conditions. Here we introduce and implement a physics-driven contrastive learning scheme for a network of variable resistors, using circuitry to locally compare the response of two identical networks subjected to the two different sets of boundary conditions. Using this innovation, our system effectively trains itself, optimizing its resistance values without use of a central processor or external information storage. Once the system is trained for a specified allostery, regression, or classification task, the task is subsequently performed rapidly and automatically by the physical imperative to minimize power dissipation in response to the given voltage inputs. We demonstrate that, unlike typical computers, such learning systems are robust to extreme damage (and thus manufacturing defects) due to their decentralized learning. Our twin-network approach is therefore readily scalable to extremely large or nonlinear networks where its distributed nature will be an enormous advantage; a laboratory network of only 500 edges will already outpace its in silico counterpart.

cond-mat.dis-nn↗

arXiv2020

Beyond Quality and Quantity: Contact Distribution Encodes Frictional Strength

Classically, the quantity of contact area $A_R$ between two bodies is considered a proxy for the force of friction. However, bond density across the interface - quality of contact - is also relevant, and contemporary debate often centers around the relative importance of these two factors. In this work, we demonstrate that a third factor, often overlooked, plays a significant role in static frictional strength: the distribution of contact. We perform static friction measurements, $\mu$, on three pairs of solid blocks while imaging the contact plane. By using linear regression on hundreds of image-$\mu$ pairs, we are able to predict future friction measurements with 3 to 7 times better accuracy than existing benchmarks, including total quantity of contact area. Our model has no access to quality of contact, and we therefore conclude that a large portion of the interfacial state is encoded in the spatial distribution of contact, rather than its quality or quantity.

cond-mat.soft↗

arXiv2019

Influences of microcontact shape on the state of a frictional interface

The real area of contact of a frictional interface changes rapidly when the normal load is altered, and evolves slowly when normal load is held constant, aging over time. Traditionally, the total area of contact is considered a proxy for the frictional strength of the interface. Here we show that the state of a frictional interface is not entirely defined by the total real area of contact but depends on the geometrical nature of that contact as well. We directly visualize an interface between rough elastomers and smooth glass and identify that normal loading and frictional aging evolve the interface differently, even at a single contact level. We introduce a protocol wherein the real area of contact is held constant in time. Under these conditions, the interface is continually evolving; small contacts shrink and large contacts coarsen.

cond-mat.soft↗

arXiv2019

Shear Destruction of Frictional Aging and Memory

We simultaneously measure the static friction and the real area of contact between two solid bodies. Under static conditions both quantities increase logarithmically in time, a phenomenon coined aging. Indeed, frictional strength is traditionally considered equivalent to the real area of contact. Here we show that this equivalence breaks down when a static shear load is applied during aging. The addition of such a shear load accelerates frictional aging while the aging rate of the real area of contact is unaffected. Moreover, a negative static shear - pulling instead of pushing - slows frictional aging, but similarly does not affect the aging of contacts. The origin of this shear effect on aging is geometrical. When shear load is increased, minute relative tilts between the two blocks prematurely erase interfacial memory prior to sliding, negating the effect of aging. Modifying the loading point of the interface eliminates these tilts and as a result frictional aging rate becomes insensitive to shear. We also identify a secondary memory-erasure effect that remains even when all tilts are eliminated and show that this effect can be leveraged to accelerate aging by cycling between two static shear loads.

cond-mat.soft↗