An Evolutionary Hypothesis for White's Illusion: Constant Percentage Correction and Modern Individual Variation
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.
Ines: I'm Ines.
Marcus: And I'm Marcus, here with Yuki.
Ines: We're looking at the paper: "An Evolutionary Hypothesis for White's Illusion".
Marcus: Here's the gist of the paper. White’s illusion is a striking demonstration of context-dependent brightness and lightness perception In this paper, we propose an evolutionary hypothesis for this phenomenon The model provides a compact computational account of the observed population-level regularity and motivates a testable evolutionary interpretation of individual variation in contextual lightness judgments The Core Hypothesis We suggest that a simple constant-percentage correction function could have emerged as an adaptive mechanism for maintaining useful material and reflectance judgments under variable illumination We hypothesize that early visual systems may have developed a fast, low-complexity mechanism that approximated this compensation through a constant percentage correction based on local average luminance Such a mechanism would not require an explicit physical reconstruction of illumination and reflectance If such a mechanism became conserved, a population-level response could remain stable even when individuals exhibit modest variation in its effective parameter Computational Framework The forward perception model is defined by the equation LO = yLP + (1 − y)Ia The background-dependent derivative of this expression is ∂LO/∂Ia = 1 − y Here, the parameter y controls the relative contribution of the desired luminance and the local average illumination In this interpretation proposed here, the population-level value of y is instead viewed as a possible expression of an potentially shared constant-percentage correction mechanism Empirical Observations The experiments established individual variation together with a stable aggregate response The observations reported in connection with our earlier forward model suggest two simultaneous properties of the constant-perception settings First, participants did not choose the same correction value Second, when these individual responses were considered together, their average produced a remarkably stable response curve across the tested conditions Implications and Future Directions The present data alone cannot determine whether the stability is genetically inherited, developmentally acquired, environmentally induced, or produced by some combination of these factors The hypothesis leads naturally to further tests Cross-cultural studies could examine whether the mean correction remains stable while the distribution of individual parameters changes A decisive next step is a calibrated, participant-level evaluation of the correction parameter across BT, WT, and EQ conditions Such tests could determine both the descriptive range of the proposed rule and whether its population-level stability generalizes beyond the conditions examined here The present manuscript does not supply a new quantitative benchmark against ODOG-family models or establish the evolutionary origin of the mechanism Future investigations should examine whether a common correction structure is evolutionarily conserved, while its effective parameter is adjusted through experience and perceptual learning The authors reviewed and take full responsibility for the content of the manuscript Limitations First, the observed stability of the population-level correction requires more extensive quantitative evaluation Future studies should examine the distributions of individual correction parameters, quantify their variability, and investigate whether the approximately constant population mean persists across larger and more diverse participant groups Second, the experimental methodology could be extended through more systematic psychophysical studies Such experiments would help distinguish genuine perceptual regularities from possible effects of experimental procedures Finally, the proposed model should be quantitatively compared with established computational explanations of lightness perception, including ODOG and its variants Systematic comparisons across multiple visual illusions are essential for evaluating the generality of computational models We have proposed an evolutionary hypothesis for the stable population-level behavior observed in our constant-luminance experiments The present manuscript does not supply a new quantitative benchmark against ODOG-family models or establish the evolutionary origin of the mechanism Rather, they identify a computational regularity and provide a biologically motivated, testable explanation for it The observed combination of individual differences and an approximately stable population-level correction is compatible with empirical accounts of visual perception The present study places the proposed constant-percentage correction in a broader set of White’s illusion configurations The observed combination of individual differences and an approximately stable population-level correction is compatible with empirical accounts of visual perception The present manuscript does not supply a new quantitative benchmark against ODOG-family models or establish the evolutionary origin of the mechanism The authors reviewed and take full responsibility for the content of the manuscript References Ergun Akleman and Tevfik Akgun.
Ines: Constant luminance perception program Barbara Blakeslee, Davis Cope, and Mark E.
Marcus: McCourt.
Ines: The oriented difference of gaussians (odog) model of brightness perception: Overview and executable mathematica notebooks Predrag Nedimovic, Sun ´ cica Zdravkovi ˇ c, and Dražen Domijan.
Paper summary: Ines: We’ve been looking at the paper "An Evolutionary Hypothesis for White's Illusion: Constant Percentage Correction and Modern Individual Variation." The authors propose that a simple constant percentage correction function could have emerged as an adaptive mechanism for maintaining useful material and reflectance judgments under variable illumination.
Marcus: They connect this to their computational model where the population-level value of Y is seen as that potential shared constant percentage correction mechanism.
Yuki: What’s the big picture here? It seems to suggest that the stability we see in human perception, even with individual differences in how much correction each person uses, is an expression of a mechanism that could have been useful for early visual systems when they needed to recognize surfaces under shadows or changing illumination.
Ines: Essentially, they are connecting the observed stability of the mean response to this simple mechanism that could have been useful for visual judgments under variable illumination, which is what makes this paper interesting.
Marcus: It’s a computationally explicit hypothesis that links the stability of the mean response to a simple function, which isn't just describing an illusion but suggesting an adaptive biological function.
Yuki: The implication for the wider field is that we might be looking at something that is evolutionarily conserved, even if its effective parameter Y gets adjusted through experience and perceptual learning.
Ines: The paper also flags a limitation, though, because they admit the observed stability of the population-level correction requires more extensive quantitative evaluation. They need to examine the distributions of individual correction parameters and quantify their variability across larger groups.
Marcus: And they also point out that future work should quantitatively compare this proposed model with established computational explanations like ODOG family models to see how general it is.
Yuki: So, while the hypothesis is compelling because it’s biologically motivated, the next step isn't just accepting it as true; it’s rigorously testing if that constant percentage correction structure holds up across diverse populations and different conditions.
Conclusion: Ines: So, to wrap up this part of our talk, they're putting forward this evolutionary hypothesis for White's illusion—that simple constant percentage correction function might have been an adaptive mechanism for handling variable light.
Marcus: Yeah, it’s about taking those individual differences we saw in the experiments and framing them as a possible sign that a shared rule evolved.
Yuki: From my side, this makes me think about how these perceptual rules spread through populations and how they might have been shaped by selective pressures over time.
Ines: Exactly. They aren't just describing *what* happens in the illusion; they’re suggesting *why* the population behaves the way it does in terms of that correction factor.
Marcus: The authors are really trying to bridge that gap between what we measure in a lab and how these systems might have started out biologically.
Yuki: It shifts our focus from just individual variation to how those variations get maintained across generations if they serve some kind of adaptive function.
Ines: And the structure of the paper suggests they’re not claiming this is proven, but rather offering a testable framework for what we should look for next.
Marcus: Right, it’s an evolutionary hypothesis, which means it’s a starting point—a computational story—that needs real data to back up the biological claims.
Yuki: So the big question becomes whether this simple constant correction actually persists when you look at how different groups of people respond across different cultures and environments.
Ines: That's what they are pushing for next: testing if that mean correction stays stable while the way people vary around it changes.
Yeditepe University · Texas A&M University
q-bio.NC, cs.CV
Submitted: 2026-10-23
Updated: 2026-10-23
Comments: 12 pages and 7 figures. We may later send a shorter version to a conference
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 80/100
The gist: White’s illusion is a striking demonstration of context-dependent brightness and lightness perception In this paper…
Key concepts
- Constant Percentage Correction
- This is a proposed adaptive mechanism where the visual system applies a fixed percentage adjustment based on local average brightness. Instead of needing to physically reconstruct the light source, this simple rule helps maintain consistent material and reflectance judgments even when illumination changes.
- Forward Perception Model
- This mathematical equation (LO = yLP + (1 − y)Ia) describes how a person perceives lightness (LO). The parameter 'y' in this model is interpreted as representing the population-level value of the proposed constant-percentage correction mechanism.
- Population-Level Stability
- The study observed that while individual participants chose different correction values, their average response curve remained remarkably stable across tested conditions. This stability suggests that the underlying mechanism—the constant percentage correction—is a conserved feature in the population.
- Individual Variation
- Experiments showed that participants did not all choose the same correction value. This individual difference is key; the hypothesis suggests this variation arises from experience or learning, but the overall population behavior remains predictable due to the shared underlying rule.
Terminology
Summary
White’s illusion is a striking demonstration of context-dependent brightness and lightness perception In this paper, we propose an evolutionary hypothesis for this phenomenon The model provides a compact computational account of the observed population-level regularity and motivates a testable evolutionary interpretation of individual variation in contextual lightness judgments
The Core Hypothesis
We suggest that a simple constant-percentage correction function could have emerged as an adaptive mechanism for maintaining useful material and reflectance judgments under variable illumination We hypothesize that early visual systems may have developed a fast, low-complexity mechanism that approximated this compensation through a constant percentage correction based on local average luminance Such a mechanism would not require an explicit physical reconstruction of illumination and reflectance If such a mechanism became conserved, a population-level response could remain stable even when individuals exhibit modest variation in its effective parameter
Computational Framework
The forward perception model is defined by the equation LO = yLP + (1 − y)Ia The background-dependent derivative of this expression is ∂LO/∂Ia = 1 − y Here, the parameter y controls the relative contribution of the desired luminance and the local average illumination In this interpretation proposed here, the population-level value of y is instead viewed as a possible expression of an potentially shared constant-percentage correction mechanism
Empirical Observations
The experiments established individual variation together with a stable aggregate response The observations reported in connection with our earlier forward model suggest two simultaneous properties of the constant-perception settings First, participants did not choose the same correction value Second, when these individual responses were considered together, their average produced a remarkably stable response curve across the tested conditions
Implications and Future Directions
The present data alone cannot determine whether the stability is genetically inherited, developmentally acquired, environmentally induced, or produced by some combination of these factors The hypothesis leads naturally to further tests Cross-cultural studies could examine whether the mean correction remains stable while the distribution of individual parameters changes A decisive next step is a calibrated, participant-level evaluation of the correction parameter across BT, WT, and EQ conditions Such tests could determine both the descriptive range of the proposed rule and whether its population-level stability generalizes beyond the conditions examined here The present manuscript does not supply a new quantitative benchmark against ODOG-family models or establish the evolutionary origin of the mechanism Future investigations should examine whether a common correction structure is evolutionarily conserved, while its effective parameter is adjusted through experience and perceptual learning The authors reviewed and take full responsibility for the content of the manuscript
Limitations
First, the observed stability of the population-level correction requires more extensive quantitative evaluation Future studies should examine the distributions of individual correction parameters, quantify their variability, and investigate whether the approximately constant population mean persists across larger and more diverse participant groups Second, the experimental methodology could be extended through more systematic psychophysical studies Such experiments would help distinguish genuine perceptual regularities from possible effects of experimental procedures Finally, the proposed model should be quantitatively compared with established computational explanations of lightness perception, including ODOG and its variants Systematic comparisons across multiple visual illusions are essential for evaluating the generality of computational models
Conclusion
We have proposed an evolutionary hypothesis for the stable population-level behavior observed in our constant-luminance experiments The present manuscript does not supply a new quantitative benchmark against ODOG-family models or establish the evolutionary origin of the mechanism Rather, they identify a computational regularity and provide a biologically motivated, testable explanation for it The observed combination of individual differences and an approximately stable population-level correction is compatible with empirical accounts of visual perception The present study places the proposed constant-percentage correction in a broader set of White’s illusion configurations The observed combination of individual differences and an approximately stable population-level correction is compatible with empirical accounts of visual perception The present manuscript does not supply a new quantitative benchmark against ODOG-family models or establish the evolutionary origin of the mechanism The authors reviewed and take full responsibility for the content of the manuscript
References
Ergun Akleman and Tevfik Akgun. Constant luminance perception program Barbara Blakeslee, Davis Cope, and Mark E. p. 7 Our specific question is narrower: whether the stable mean observed in our constant-perception
experiments can be explained by a conserved constant-percentage correction mechanism, while allowing its effective
parameter to vary among individuals. 4 Forward Model
As in our previous work, let LP denote the target luminance parameter used to specify the intended appearance, LO the
computed display luminance, and Ia a local average of the displayed background luminance (not a physical estimate of
incident illumination). Established computational explanations of White’s illusion. Oriented Difference-of-Gaussians (ODOG) filtering
with normalization provides a prominent mechanistic explanation for White’s effect and related brightness phenomena Blakeslee et al. Corney and Lotto Corney and
Lotto [2007] demonstrated that White-like illusions can emerge from learning about reflectance in three-dimensional
environments, while Purves and colleagues Purves et al. [2015] emphasized the role of empirical experience in visual
perception. Ecological and empirical accounts. Corney and Lotto trained networks to infer surface reflectance in simulated
three-dimensional scenes and reported that White-like assimilation emerged with three-dimensional rather than twodimensional training environments Corney and Lotto [2007]. The model
therefore provides a compact computational account of the observed population-level regularity
and motivates a testable evolutionary interpretation of individual variation in contextual lightness
judgments. 1 Introduction
White’s illusion, like many classical lightness illusions, reveals that perceived brightness and lightness depend strongly
on context. Their result suggests that scene statistics and learned
inference can generate perceptual effects without prescribing our specific blending rule. Purves and colleagues likewise
argue that perceptual responses may reflect the empirical utility of prior experience rather than recovery of physical
scene properties Purves et al. 2
1
Some researchers prefer the term “lightness perception” Gilchrist et al. We use luminosity perception when discussing
the broader physical and computational setting, while recognizing the established terminology of lightness perception in the vision
literature. Experience,
calibration, decision strategy, and task-dependent response noise are all plausible contributors. For example, probability
matching has been documented as a perceptual decision strategy in a multisensory task Wozny et al. Experience,
calibration, decision strategy, and task-dependent response noise are all plausible contributors. For example, probability
matching has been documented as a perceptual decision strategy in a multisensory task Wozny et al. (e) BT 200. p. 2 (d) BT 150. (e) BT 200. A central difficulty is that the luminance measured at the eye does not
uniquely determine surface reflectance: illumination, spatial arrangement, surface orientation, incident-light direction,
and occlusion can all affect the observed signal Gilchrist [1977, 1980], Gilchrist et al. Related psychophysical
studies likewise demonstrate that perceived brightness and luminosity depend strongly on surrounding luminance and
adaptation Yarbus and Yarbus [1967], Whittle and Challands [1969], Shapley and Enroth-Cugell [1984], Bonato and
Gilchrist [1994]. A central difficulty is that the luminance measured at the eye does not
uniquely determine surface reflectance: illumination, spatial arrangement, surface orientation, incident-light direction,
and occlusion can all affect the observed signal Gilchrist [1977, 1980], Gilchrist et al. Related psychophysical
studies likewise demonstrate that perceived brightness and luminosity depend strongly on surrounding luminance and
adaptation Yarbus and Yarbus [1967], Whittle and Challands [1969], Shapley and Enroth-Cugell [1984], Bonato and
Gilchrist [1994]. A central difficulty is that the luminance measured at the eye does not
uniquely determine surface reflectance: illumination, spatial arrangement, surface orientation, incident-light direction,
and occlusion can all affect the observed signal Gilchrist [1977, 1980], Gilchrist et al. Related psychophysical
studies likewise demonstrate that perceived brightness and luminosity depend strongly on surrounding luminance and
adaptation Yarbus and Yarbus [1967], Whittle and Challands [1969], Shapley and Enroth-Cugell [1984], Bonato and
Gilchrist [1994]. A central difficulty is that the luminance measured at the eye does not
uniquely determine surface reflectance: illumination, spatial arrangement, surface orientation, incident-light direction,
and occlusion can all affect the observed signal Gilchrist [1977, 1980], Gilchrist et al. Related psychophysical
studies likewise demonstrate that perceived brightness and luminosity depend strongly on surrounding luminance and
adaptation Yarbus and Yarbus [1967], Whittle and Challands [1969], Shapley and Enroth-Cugell [1984], Bonato and
Gilchrist [1994]. In earlier work, we developed a forward translucency-based formula capable of producing
approximately constant perceived appearance across varying backgrounds. While validating that
formulation, we observed a notable population-level pattern: individuals selected slightly different
correction percentages for “constant perception,” while the population average remained remarkably
stable across experiments. 1
A computationally related problem appears in intrinsic image analysis, where an observed image is interpreted in terms
of underlying scene properties such as reflectance and illumination. Modern intrinsic-image methods can employ
optimization, depth information, learned priors, user input, or data-driven decomposition Chen and Koltun [2013], Shen
et al. Modern intrinsic-image methods can employ
optimization, depth information, learned priors, user input, or data-driven decomposition Chen and Koltun [2013], Shen
et al. [2013], Li and Snavely [2018], and evaluation itself requires suitable ground truth Grosse et al. Modern intrinsic-image methods can employ
optimization, depth information, learned priors, user input, or data-driven decomposition Chen and Koltun [2013], Shen
et al. [2013], Li and Snavely [2018], and evaluation itself requires suitable ground truth Grosse et al. [2013], Li and Snavely [2018], and evaluation itself requires suitable ground truth Grosse et al. Despite
substantial progress, intrinsic decomposition remains a challenging computer-vision problem Garces et al. Despite
substantial progress, intrinsic decomposition remains a challenging computer-vision problem Garces et al. Our
approach differs in purpose: we do not propose that biological vision explicitly constructs intrinsic images. A biologically useful mechanism need not
recover physically correct reflectance or illumination; it need only improve behavior under variable viewing conditions. Ecological pressures associated with locating food and recognizing biologically relevant objects provide examples
of how visual discrimination can be shaped by natural environments Regan et al. At the same time, an evolutionary account of a basic mechanism does not imply that all visual responses must be
identical across populations or environments. Research on cultural and environmental influences shows that contextual
processing, field dependence, experience with built environments, and depth cues can vary across human groups and
experience Nisbett and Miyamoto [2005], Ji et al. [2000], Segall et al. [1966], Deregowski [1980].Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 1 ---
AN EVOLUTIONARY HYPOTHESIS FOR WHITE’S ILLUSION:
CONSTANT PERCENTAGE CORRECTION AND MODERN
INDIVIDUAL VARIATION
A PREPRINT
Bekir Tevfik Akgün Faculty of Computer and Information Sciences Yeditepe University, Turkey bekirtevfik.akgun@yeditepe. p. 0 AN EVOLUTIONARY HYPOTHESIS FOR WHITE’S ILLUSION:
CONSTANT PERCENTAGE CORRECTION AND MODERN
INDIVIDUAL VARIATION
A PREPRINT
Bekir Tevfik Akgün
Faculty of Computer and Information Sciences
Yeditepe University, Turkey
bekirtevfik. akgun@yeditepe. tr
Ergun Akleman
Visual Computing & Computational Media, SPVFA
Computer Science and Engineering, COE
Texas A&M University, USA
ergun@tamu. edu
November 6, 2026
ABSTRACT
White’s illusion is a striking demonstration of context-dependent brightness and lightness perception.Where the paper says it
Where the paper says it
--- Page 2 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT Figure 1: Illustration of the constant-percentage correction underlying the proposed model. The same correction percentage is applied across different local background luminances, providing the population-level reference behavior considered in this paper
--- Page 3 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT (a) BT 000. (b) BT 050. p. 2 An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT
(a) BT 000. (b) BT 050. (b) BT 050. (c) BT 100. (c) BT 100. (d) BT 150.Where the paper says it
Where the paper says it
Where the paper says it
--- Page 4 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT (a) BT050165165. (b) BT0501693
--- Page 5 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT (a) EQ 000. (b) EQ 050. p. 2 An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT
(a) BT 000. (b) BT 050. (b) BT 050. (c) BT 100. (c) BT 100. (d) BT 150. (d) BT 150. (e) BT 200.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 6 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT (a) EQ050165165. (b) EQ0501693
--- Page 7 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT (a) WT 000. (b) WT 050. p. 2 An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT
(a) BT 000. (b) BT 050. (b) BT 050. (c) BT 100. (c) BT 100. (d) BT 150. (d) BT 150. (e) BT 200.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 8 ---
An Evolutionary Hypothesis for White’s Illusion:
Constant Percentage Correction and Modern Individual Variation A PREPRINT (a) WT050165165.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Develop a computational model that predicts perceived appearance by applying a constant percentage correction to background luminance, as proposed in Equation 1 (LO = yLP + (1 − y)Ia). This allows AI systems to generate stimuli that maintain an approximately constant appearance across varying backgrounds, which is crucial for simulating context-dependent brightness perception.
-
Implement a mechanism where individual observers exhibit variation in the correction parameter y while maintaining a stable population mean, reflecting the hypothesis:
individual differences may reflect learning, adaptation, or variation around a shared computational response.
This enables AI to model how conserved mechanisms can coexist with individual parameter adjustments. -
Create an evolutionary hypothesis framework for visual systems by testing whether the observed stability of the population mean can be explained by a
conserved perceptual mechanism that may have been useful for maintaining visual judgments under changing illumination.
This moves AI beyond mere description toward proposing functional, adaptive origins for perceptual rules. -
Enhance intrinsic image analysis algorithms to prioritize a
simple local correction
over explicit physical reconstruction of illumination and reflectance, aligning with the paper's distinction:we do not propose that biological vision explicitly constructs intrinsic images.
This allows AI to focus on economically useful approximations for perceptual constancy rather than computationally exhaustive scene decomposition. -
Integrate comparative testing protocols that examine whether the
stability of the population mean reflects evolutionary conservation, environmental learning, or a combination of these influences
across different cultures and age groups. This capability allows AI systems to evaluate the robustness and generalization of perceptual rules beyond a single experimental context.
Abstract
White's illusion is a striking demonstration of context-dependent brightness and lightness perception. In earlier work, we developed a forward translucency-based formula capable of producing approximately constant perceived appearance across varying backgrounds. While validating that formulation, we observed a notable population-level pattern: individuals selected slightly different correction percentages for "constant perception," while the population average remained remarkably stable across experiments. In this paper, we propose an evolutionary hypothesis for this phenomenon. We suggest that a simple constant-percentage correction function could have emerged as an adaptive mechanism for maintaining useful material and reflectance judgments under variable illumination. Under this interpretation, the stability of the population mean may reflect a conserved perceptual mechanism, whereas individual differences may reflect learning, adaptation, or variation around a shared computational response. Using a modification of our previously published forward model, we show that White's illusion can be reproduced by applying a constant-percentage correction to background luminance. The model therefore provides a compact computational account of the observed population-level regularity and motivates a testable evolutionary interpretation of individual variation in contextual lightness judgments.
Related papers
- Self-Supervised Learning of Local Predictive Directions as a Neural Computation
- Synchronized Hearts, Desynchronized Brains: A Nonlinear Dynamics Perspective for Function and Failure
- Message interaction dynamics covary with brain volume across mammalian connectomes
- Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation
- The Platonic brain bridge hypothesis: brain networks as architectural priors for multimodal large language models
- BrainPFN: Amortized Brain-Behaviour Prediction via Prior-Data Fitted Networks