EARS-UDE : Evaluating Auditory Response in Sensory Overload with Universal Differential Equations

Miheer Salunke Prathamesh Dinesh Joshi Sydney, Australia Vizuara AI Labs [email protected] [email protected] ORCID: 0009-0006-7551-4147 ORCID: 0009-0008-5476-9962 Raj Abhijit Dandekar Rajat Dandekar Vizuara AI Labs Vizuara AI Labs [email protected] [email protected] Sreedath Panat · 2025

Read EARS-UDE : Evaluating Auditory Response in Sensory Overload with Universal Differential Equations by Miheer Salunke Prathamesh Dinesh Joshi Sydney, Australia Vizuara AI Labs [email protected] [email protected] ORCID: 0009-0006-7551-4147 ORCID: 0009-0008-5476-9962 Raj Abhijit Dandekar Rajat Dandekar Vizuara AI Labs Vizuara AI Labs [email protected] [email protected] Sreedath Panat (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-26804.txt.

Introduction

Miheer Salunke and four co-authors model hearing overload in autism in a 2025 study. Sound overload is when noise feels too strong and hard to bear. It affects many autistic people, yet past models often fixed traits that vary by person.

Their method, called a universal differential equation, mixes known body rules with a learned part. A differential equation is a rule that links change over time to current state. The mix beat a pure learned model and used far fewer free parts. It also found body traits with small error and flagged pulse patterns with risk. The goal is care and wearables shaped to each person.

Catalog record

CREATORMiheer Salunke Prathamesh Dinesh Joshi Sydney, Australia Vizuara AI Labs [email protected] [email protected] ORCID: 0009-0006-7551-4147 ORCID: 0009-0008-5476-9962 Raj Abhijit Dandekar Rajat Dandekar Vizuara AI Labs Vizuara AI Labs [email protected] [email protected] Sreedath Panat
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-26804

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

EARS-UDE : Evaluating Auditory Response in Sensory Overload with Universal Differential Equations Miheer Salunke Prathamesh Dinesh Joshi Sydney, Australia Vizuara AI Labs [email protected] [email protected] ORCID: 0009-0006-7551-4147 ORCID: 0009-0008-5476-9962 Raj Abhijit Dandekar Rajat Dandekar Vizuara AI Labs Vizuara AI Labs [email protected] [email protected] Sreedath Panat Vizuara AI Labs [email protected] Abstract Auditory sensory overload affects 50-70% of individuals with Autism Spectrum Disorder (ASD), yet existing approaches, such as mechanistic models (Hodgkin-Huxley-type, Wilson-Cowan, excitation-inhibition balance), clinical tools (EEG/MEG, Sensory Profile scales), and ML methods (Neural ODEs, predictive coding), either assume fixed parameters or lack interpretability, missing autism’s heterogeneity. We present a Scientific Machine Learning approach using Universal Differential Equations (UDEs) to model sensory adaptation dynamics in autism. Our framework combines ordinary differential equations grounded in biophysics with neural networks to capture both mechanistic understanding and individual variability. We demonstrate that UDEs achieve a 90.8% improvement over pure Neural ODEs while using 73.5% fewer parameters. The model successfully recovers physiological parameters within the 2% error and provides a quantitative risk assessment for sensory overload, predicting 17.2% risk for pulse stimuli with specific temporal patterns. This framework establishes foundations for personalized, evidence-based interventions in autism, with direct applications to wearable technology and clinical practice. 1 Introduction Autism Spectrum Disorder affects approximately 1-2% of the global population, with sensory processing differences now recognized as a core diagnostic feature in DSM-5 [ 1 , 2 , 3 ] . Among these differences, auditory hypersensitivity is particularly challenging, affecting 50-70% of autistic individuals and significantly affecting their ability to navigate daily environments [ 4 , 5 , 6 ] . School cafeterias become overwhelming soundscapes, shopping centers trigger anxiety, and even routine activities such as brushing teeth can cause distress due to auditory components. These sensory challenges often lead to social withdrawal, academic difficulties, and a reduced quality of life [ 7 , 8 ] . At the neurological level, sensory overload occurs when the nervous system receives more input than it can process effectively, leading to a cascade of physiological and behavioral responses [ 9 , 10 ] . Research suggests that in autism, typical adaptation mechanisms that "dampen" responses to continuous stimuli function differently [ 11 , 12 ] . This altered adaptation may involve differences in GABAergic inhibition, altered connectivity between sensory and regulatory brain regions, and atypical predictive coding mechanisms [ 13 , 14 , 15 ] . Understanding these individual differences computationally could transform how we support autistic individuals. Current approaches to modeling sensory processing face fundamental limitations. Classical mechanistic models based on ordinary differential equations (ODEs) capture known neurophysiology, but assume fixed parameters across all individuals, missing the heterogeneity that characterizes autism [ 16 , 17 , 18 ] . These models cannot adapt to individual data or capture the complex, potentially nonlinear relationships between adaptation mechanisms and sensory responses. In contrast, pure machine learning approaches like Neural ODEs [ 19 , 20 ] can learn complex patterns from data but operate as black boxes, providing no mechanistic insight into the underlying processes. This lack of interpretability prevents clinical translation, as practitioners cannot understand why certain interventions might work for specific individuals [ 21 , 22 ] . We address this challenge through Universal Differential Equations [ 23 , 24 ] , a Scientific Machine Learning framework that optimally combines mechanistic understanding with data-driven flexibility. Rather than choosing between interpretability and accuracy, UDEs augment known physiological structures with neural networks only where our knowledge is incomplete [ 25 , 26 ] . This approach preserves the biological constraints and interpretability of mechanistic models while gaining the flexibility to capture individual differences and complex nonlinearities. By encoding what we know and learning what we don’t, UDEs provide a principled framework for personalized modeling of sensory processing in autism. The paper is organized as follows. Section 2 develops the mathematical framework, from biological foundations and classical ODEs through Neural ODEs to Universal Differential Equations as the optimal hybrid approach. Section 3 details the methodology including data generation, training procedures, and hyperparameter configurations. Section 4 presents results demonstrating 90.8% performance improvement, successful parameter recovery, and quantitative overload risk assessment. Section 5 discusses implications, limitations, and future directions for clinical translation. 2 Mathematical Framework for Sensory Adaptation 2.1 Biological Foundations and Classical ODE Model Sensory adaptation involves multiple neurophysiological processes that work in concert. At the receptor level, sustained stimulation leads to decreased sensitivity through phosphorylation and conformational changes [ 30 , 31 ] . In synapses, vesicle depletion and receptor desensitization reduce signal transmission [ 32 , 33 ] . In neural circuits, inhibitory interneurons provide feedback control, while higher-level predictive coding mechanisms adjust expectations according to context [ 27 , 34 , 35 ] . These processes operate on different timescales, milliseconds for synaptic effects, seconds for adaptation at the circuit level, and minutes for long-term habituation [ 36 , 37 ] . These equations are derived from established neurophysiological principles of sensory adaptation [ 31 , 36 ] , synaptic depression models [ 33 , 32 ] , and cortical gain control mechanisms [ 42 ] : \displaystyle\frac{dR}{dt} \displaystyle=\alpha S(t)-\beta A(t)R(t) (1) \displaystyle\frac{dA}{dt} \displaystyle=\gamma R(t)-\delta A(t) (2) Here, R(t) represents the neurological response corresponding to the aggregate firing rates in the auditory cortex, measurable through EEG or MEG recordings [ 38 , 39 ] . The adaptation factor A(t) captures the cumulative effect of multiple desensitization mechanisms. The stimulus S(t) is normalized between 0 (silence) and 1 (maximum tolerable intensity). The parameter \alpha determines the response sensitivity of individuals with higher \alpha who experience stronger initial responses to stimuli, which could reflect differences in excitatory/inhibitory balance [ 40 , 41 ] . The effectiveness of adaption \beta controls how strongly adaptation suppresses responses, possibly related to the strength of GABAergic inhibition. The build-up rate \gamma governs the speed with which adaptation develops, while the decay rate \delta determines the recovery speed after stimulus removal. The multiplicative interaction A(t)R(t) in Equation 1 captures a key biological principle, the effectiveness of adaptation scales with the magnitude of the response. Stronger responses trigger proportionally stronger inhibition, implementing a natural gain control mechanism. This nonlinearity, observed in neural recordings, provides protection against overwhelming stimulation when functioning properly [ 42 , 43 ] , as shown in Figure 1 . (a) Ground truth dynamics without noise (b) Realistic data with measurement noise Figure 1: Classical ODE model of sensory adaptation. (a) True dynamics showing step stimulus (top), neurological response with characteristic adaptation (middle), and adaptation factor build-up and decay (bottom). (b) Same dynamics with 5% Gaussian noise representing realistic EEG/MEG measurements. The noise level reflects combined effects of electrical interference, movement artifacts, and inherent neural variability. 2.2 Limitations of Fixed-Parameter Models Although mechanistically grounded, classical ODEs make restrictive assumptions that limit clinical utility. Parameters are assumed constant across individuals, yet autism research consistently shows enormous heterogeneity in sensory profiles [ 44 , 45 ] . Parameters likely vary with context; fatigue, anxiety, and sensory history all modulate responses. The simple linear term \beta A(t) cannot capture the potential saturation effects, threshold phenomena, or state-dependent modulation observed in neural systems [ 46 , 47 ] . Most importantly, there is no systematic way to determine individual parameters from the observed

Read the complete text