arXiv subjects

Manoj Srinivasan

Publications and source records attributed to Manoj Srinivasan.

7 recordsLinked to original sources

arXiv2026

Paradoxical noise preference in RNNs

In recurrent neural networks (RNNs) used to model biological neural networks, noise is typically introduced during training to emulate biological variability and regularize learning. The expectation is that removing the noise at test time should preserve or improve performance. Contrary to this intuition, we find that continuous-time RNNs (CTRNNs) often perform best at or near the training noise level. This noise preference typically arises when noise is injected inside the neural activation function; networks trained with noise injected outside the activation function perform best with zero noise. The phenomenon arises robustly in diverse tasks for large enough training noise; we also show the phenomenon arising in feedforward neural networks, not just in RNNs. Our analyses show that the phenomenon stems from noise-induced shifts of fixed points (stationary distributions) in the underlying stochastic dynamics of the RNNs. These fixed point shifts are noise-level dependent and bias the network outputs when the noise is removed, degrading performance. Analytical and numerical results show that the bias arises when neural states operate near activation-function nonlinearities, where noise is asymmetrically attenuated, and that performance optimization incentivizes operation near these nonlinearities; such performance incentives exist for networks with noise inside, but not outside, the activation function, explaining why only noise-in networks show the preference. Thus, networks can overfit to the training noise itself rather than just to the input-output data. The phenomenon is distinct from stochastic resonance, wherein nonzero noise enhances signal processing. Our findings reveal that training noise can become an integral part of the computation learned by neural networks, with implications for understanding neural population dynamics and for the design of robust artificial RNNs.

cs.NE↗

arXiv2025

Overground gait transitions are not sharp but involve gradually changing walk-run mixtures even over long distances

Humans typically walk at low speeds and run at higher speeds. Previous studies of transitions between walking and running were mostly on treadmills, but real-world locomotion allows more flexibility. Here, we study overground locomotion over long distances (800 m or 2400 m) under time constraints, simulating everyday scenarios like going to an appointment. Unlike on treadmills, participants can vary both speed and gait during this task. We find that gait transition in this overground task occurs over a broad `gait transition regime' spanning average speeds from 1.9 m/s to 3.0 m/s. In this regime, people use mixtures of walking and running: mostly walking at low average speeds (around 1.9 m/s) and mostly running at high average speeds (3.0 m/s); the walk vs run fraction gradually changes between these speed limits. Within any walk-run mixture, there is a speed gap between the walking and running. These gait mixtures and their specific structure are predicted by energy optimality. These findings extend earlier results from much shorter distance tasks, showing that similar energetic principles govern longer, more physically and cognitively demanding tasks. Overall, our results highlight the role of whole-task energy minimization including transients in shaping human locomotion.

q-bio.NC↗

arXiv2025

Metabolic energy expenditure for time-varying isometric forces

Muscles consume metabolic energy (ATP) to produce force. A mathematical model for energy expenditure can be useful in estimating real-time costs of movements or to predict energy optimal movements. Metabolic cost models developed so far have predominantly aimed at dynamic movement tasks, where mechanical work dominates. Further, while it is known that both force magnitude and rate of change of force (force rate) affect metabolic cost, it is not known how these terms interact, or if the force rate dependence can be a consequence of the force dependence. Here, we performed extensive human subject experiments, involving each subject over 5 hours of metabolic trials, which systematically changed the mean forces and forces rates so as to characterize a holistic relation for metabolic cost based on both force and force rate -- or analogously, torque and torque rate. Our experiments involved humans producing symmetric or asymmetric sinusoidal forces with different means, amplitudes, frequencies, and rise and fall periods. We showed that the metabolic cost can be well-approximated by a sum of power law functions of torque and torque rate. We found that the metabolic cost scales non-linearly with joint torque (with exponent = 1.36) and non-linearly with torque rate (with exponent = 2.5). Surprisingly, the data suggested that the cost was roughly four times higher for decreasing the torque than increasing, mirroring the analogous ratio between the cost of positive and negative work. Using these metabolic cost relations, we show that if the metabolic cost scales with particular exponents with muscle force and force rates, the same exponents will be observed in multi-joint tasks with multiple muscles. Our new metabolic cost model involving both force and force rate will potentially allow better predictions of energy optimal movements and thus inform wearable robot design and analysis.

q-bio.TO↗

arXiv2022

Walking for short distances and turning in lower-limb amputees: a study in low-cost prosthesis users

Preferred walking speed is a widely-used performance measure for people with mobility issues, but is usually measured in straight line walking for fixed distances or durations. However, daily walking involves walking for bouts of different distances and walking with turning. Here, we studied walking for short distances and walking in circles in unilateral lower-limb amputees wearing an above or below-knee passive prosthesis, specifically, a Jaipur foot prosthesis. Analogous to earlier results in non-amputees, we found that their preferred walking speeds are lower for short distances and lower for circles of smaller radii. Using inverse optimization, we estimated the cost of changing speeds and turning such that the observed preferred walking speeds in our experiments minimizes the total energy cost. The inferred costs of changing speeds and turning were much larger than for non-amputees. These findings could inform prosthesis design and rehabilitation therapy to better assist changing speeds and turning tasks in amputee walking. Further, measuring the preferred speed for a range of distances and radii is a more robust subject-specific measure of walking performance.

q-bio.QM↗

arXiv2022

Towards skin-acetone monitors with selective sensitivity: dynamics of PANI-CA films

Most research aimed at measuring biomarkers on the skin is only concerned with sensing chemicals in sweat using electrical signals, but these methods are not truly non-invasive nor non-intrusive because they require substantial amounts of sweat to get a reading. This project aims to create a truly non-invasive wearable sensor that continuously detects the gaseous acetone (a biomarker related to metabolic disorders) that ambiently comes out of the skin. Composite films of polyaniline and cellulose acetate, exhibiting chemo-mechanical actuation upon exposure to gaseous acetone, were tested in the headspaces above multiple solutions containing acetone, ethanol, and water to gauge response sensitivity, selectivity, and repeatability. The bending of the films in response to exposures to these environments was tracked by an automatic video processing code, which was found to out-perform an off-the-shelf deep neural network-based tracker. Using principal component analysis, we showed that the film bending is low dimensional with over 90% of the shape changes being captured with just two parameters. We constructed forward models to predict shape changes from the known exposure history and found that a linear model can explain 40% of the observed variance in film tip angle changes. We constructed inverse models, going from third order fits of shape changes to acetone concentrations where about 45% of the acetone variation and about 30% of ethanol variation are captured by linear models, and non-linear models did not perform substantially better. This suggests there is sufficient sensitivity and inherent selectivity of the films. These models, however, provide evidence for substantial hysteretic or long-time-scale responses of the PANI films, seemingly due to the presence of water. Further experiments will allow more accurate discrimination of unknown exposure environments.

physics.med-ph↗

arXiv2021

A unified energy optimality criterion predicts human navigation paths and speeds

Navigating our physical environment requires changing directions and turning. Despite its ecological importance, we do not have a unified theoretical account of non-straight-line human movement. Here, we present a unified optimality criterion that predicts disparate non-straight-line walking phenomena, with straight-line walking as a special case. We first characterized the metabolic cost of turning, deriving the cost landscape as a function of turning radius and rate. We then generalized this cost landscape to arbitrarily complex trajectories, allowing the velocity direction to deviate from body orientation (holonomic walking). We used this generalized optimality criterion to mathematically predict movement patterns in multiple contexts of varying complexity: walking on prescribed paths, turning in place, navigating an angled corridor, navigating freely with end-point constraints, walking through doors, and navigating around obstacles. In these tasks, humans moved at speeds and paths predicted by our optimality criterion, slowing down to turn and never using sharp turns. We show that the shortest path between two points is, counterintuitively, often not energy optimal, and indeed, humans do not use the shortest path in such cases. Thus, we have obtained a unified theoretical account that predicts human walking paths and speeds in diverse contexts. Our model focuses on walking in healthy adults; future work could generalize this model to other human populations, other animals, and other locomotor tasks.

q-bio.NC↗

arXiv2008

Rocking and rolling: a can that appears to rock might actually roll

A beer bottle or soda can on a table, when slightly tipped and released, falls to an upright position and then rocks up to a somewhat opposite tilt. Superficially this rocking motion involves a collision when the flat circular base of the container slaps the table before rocking up to the opposite tilt. A keen eye notices that the after-slap rising tilt is not generally just diametrically opposite the initial tilt but is veered to one side or the other. Cushman and Duistermaat (2006) recently noticed such veering when a flat disk with rolling boundary conditions is dropped nearly flat. Here, we generalize these rolling disk results to arbitrary axi-symmetric bodies and to frictionless sliding. More specifically, we study motions that almost but do not quite involve a face-down collision of the round container's bottom with the table-top. These motions involve a sudden rapid motion of the contact point around the circular base. Surprisingly, like for the rolling disk, the net angle of motion of this contact point is nearly independent of initial conditions. This angle of turn depends simply on the geometry and mass distribution but not on the moment of inertia about the symmetry axis. We derive simple asymptotic formulas for this "angle of turn" of the contact point and check the result with numerics and with simple experiments. For tall containers (height much bigger than radius) the angle of turn is just over $π$ and the sudden rolling motion superficially appears as a nearly symmetric collision leading to leaning on an almost diametrically opposite point on the bottom rim.

physics.class-ph↗