Context dependent adaptation in a neural computation
Charles J. Edelson, Sima Setayeshgar, William Bialek, Rob R. de Ruyter van Steveninck · 2025
Read Context dependent adaptation in a neural computation by Charles J. Edelson, Sima Setayeshgar, William Bialek, Rob R. de Ruyter van Steveninck (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-2509-01760.txt.
Introduction
Charles J Edelson, Sima Setayeshgar, William Bialek, and Rob R de Ruyter van Steveninck study how a fly brain cell adapts to changing light. Brains tune their response to the pattern of input they receive. In vision, brightness shifts fast, its spread shifts more slowly, and the full contrast range shifts slowest of all. The team feeds a motion sensing cell in the fly eye with natural style input split across those time scales.
The cell response to motion depends on contrast, and that link itself shifts with context. Yet the shifts live in a low dimensional space with one main axis. Across many settings, the cell tunes its averaging time to the gap between spikes. That tuning cuts redundant signals and saves coding effort.
Catalog record
| CREATOR | Charles J. Edelson, Sima Setayeshgar, William Bialek, Rob R. de Ruyter van Steveninck |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-01760
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Context dependent adaptation in a neural computation Charles J. Edelson, Sima Setayeshgar, William Bialek, Rob R. de Ruyter van Steveninck Abstract Brains adapt to the statistical structure of their input. In the visual system, local light intensities change rapidly, the variance of the intensity changes more slowly, and the dynamic range of contrast itself changes more slowly still. We use a motion–sensitive neuron in the fly visual system to probe this hierarchy of adaptation phenomena, delivering naturalistic stimuli that have been simplified to have a clear separation of time scales. We show that the neural response to visual motion depends on contrast, and this dependence itself varies with context. Using the spike–triggered average velocity trajectory as a response measure, we find that context dependence is confined to a low–dimensional space, with a single dominant dimension. Across a wide range of conditions this adaptation serves to match the integration time to the mean interval between spikes, reducing redundancy. I Introduction Adaptation is a central feature of living systems, but “adaptation” means many things. The dynamics of evolution results (in part) in adaptation by natural selection Lenski 2017 , and we describe features of the organism that enhance fitness as being adaptive. This form of adaptation has been observed over tens of thousands of generations in E. coli , but the same bacteria adapt their chemotactic behavior to constant background concentrations over the course of seconds Block et al. 1983 . We humans also adapt to constant sensory stimuli, becoming largely insensitive to absolute signals while retaining sensitivity to small variations around the constant background. Correlates of such sensory adaptation were first characterized in single neurons one hundred years ago Adrian 1928 . We now understand that sensory adaptation is more than simply subtracting a constant background Wark et al. 2007 ; Weber et al. 2019 . The natural world is complex, with its statistical characteristics shifting across both space and time. Signals can change gradually, as with the subtle shift of light intensity when summer turns to fall, or abruptly, as with the dramatic shift experienced when moving from a dense forest to an open field. Despite this complexity and variability, animals go about their daily lives, reliably responding to the signals that matter most for their survival. This simple observation helped inspire Barlow’s efficient coding hypothesis, which posits that sensory systems efficiently encode their natural stimuli Barlow 1959 ; Barlow 1961 . The key insight is that efficient coding requires sensory systems to be contextually aware and responsive to the distribution of natural stimuli. This shift toward a distributional and information-theoretic perspective provided a powerful framework for studying sensory information encoding, with visual systems serving as an effective testing ground for these ideas Laughlin 1981 ; Smirnakis et al. 1997 ; Egelhaaf and Kern 2002 ; Bialek et al. 2006 ; Wark et al. 2007 ; Kohn 2007 ; Bialek 2012 ; Webster 2015 ; Palmer et al. 2015 ; Tkačik and Bialek 2016 ; Weber et al. 2019 . 1 1 1 The idea of matching to the distribution of inputs is applied most often to strategies for coding. But when signals are noisy or ambiguous, strategies for inferring things of interest to the organism also should adapt—not just neural coding but also neural computation should match the input distribution. The computation of motion in the visual system provided an early modern example of this idea Potters and Bialek 1994 ; Weiss et al. 2002 , which has precursors reaching back to Helmholtz Barlow 1990 . These ideas expanded to a broader view of perception as Bayesian inference Knill and Richards 1996 . Fly visual systems, in particular, provide a valuable model for studying neural information encoding. The fly’s optical lobes are organized into distinct layers, each of which has been thoroughly researched, with recent studies shedding light on the complete Drosophila connectome S Dorkenwald et al. 2024 . The discovery of these well-organized structures has prompted questions regarding their functions, resulting in numerous efforts to understand how each component contributes to the fly’s ability to see and interact with the world Borst and Groschner 2023 . While substantial progress has been made overall, studies focusing on fly visual motion perception have been particularly successful. This can be at least partially attributed to the early discovery of the lobula plate tangential cells (LPTC), a collection of wide-field motion sensitive neurons located on their eponymous neural structure, the lobula plate Bishop and Keehn 1966 . These cells are directionally selective, and appear to be an important part of the fly’s flight control system, with experimental evidence for LPTC involvement in several key flight behaviors Pierantoni 1976 ; Hausen 1984 ; Krapp 2008 ; Huston and Krapp 2008 ; Ullrich et al. 2014 ; Wei et al. 2020 . Given the importance of flight to flies, it is perhaps unsurprising that LPTCs maintain robust velocity responses even in the presence of changing scene statistics. For example, H1, a spiking horizontal wide-field motion sensitive LPTC, adapts to changes in light intensity, contrast, and velocity, and in some cases to the statistics of these variations Maddess and Laughlin 1985 ; de Ruyter van Steveninck et al. 1986 ; Brenner et al. 2000a ; Fairhall et al. 2001 ; Reisenman et al. 2003 ; Flanagin 2006 . This adaptive behavior, coupled with its experimental accessibility and role in flight control, has made H1 an attractive system for studying how neurons encode information in dynamic natural environments Rieke et al. 1997 ; Brenner et al. 2000b ; Brenner et al. 2000a ; Fairhall et al. 2001 ; de Ruyter van Steveninck and Bialek 2002 ; Nemenman et al. 2008 . Adding to this experimental appeal is that visual stimuli are easy to control in a lab environment, allowing for precise manipulation of stimuli and the distributions out of which these stimuli are drawn. This comes with two major advantages. The first is the ability to ask very precisely what causes a spike Rieke et al. 1997 ; Chichilnisky 2001 . The second is the ability to repeat stimuli and measure the reliability of H1’s response de Ruyter van Steveninck et al. 1997 ; de Ruyter van Steveninck and Bialek 2002 ; Nemenman et al. 2008 . When combined, these methods can provide powerful results about the optimality and efficiency of a neural code. As an example, the dynamic range of H1’s input/output relation is scaled to the width of the input velocity distribution, maximizing information transmission Brenner et al. 2000a . In this paper, we expand upon these results and attempt to characterize how H1 adapts to both dynamic and distributional changes in the contrast of visual stimuli. Quantitatively, we characterize H1’s response to motion by the average trajectory of velocity vs time in the neighborhood of a spike—the spike triggered average de Boer and Kuyper 1968 or STA. We confirm that this response to motion depends strongly on the variance of light intensity in a scene, or contrast, and demonstrate that this dependence itself is different when contrast variations are drawn from different distributions. That is, the contrast variation of H1’s motion response depends on the range of contrast values that the fly has