Flips Reveal the Universal Impact of Memory on Random Explorations

arXiv:2506.15642 · cond-mat.stat-mech, physics.app-ph · Submitted 2025-06-18 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "Flips Reveal the Universal Impact of Memory on Random Explorations".

Mira: Here's the gist of the paper. Flips reveal the universal impact of memory on random explorations by demonstrating that in one dimension, the probability of a flip scales as 1/n independently of the underlying stochastic process.

Kai: The concept of a flip is introduced to quantify reversals in exploration, defined as transitions from a maximum to a minimum, or vice-versa, which serves as a direct measure of persistence in exploration.

Mira: While this reduces to the classical splitting probability in Markovian systems, the authors show that the flip probability follows a strikingly simple and universal law: it decays inversely with the number of sites visited, as 1/n, independently of the underlying stochastic process.

Kai: This universality arises because flips occur when traversing from one edge of its own dynamically explored region to the other, introducing memory and aging into the observable itself.

Paper summary: Kai: So we're looking at this paper called "Flips Reveal the Universal Impact of Memory on Random Explorations". It tackles how memory affects random exploration, specifically in one dimension. The main claim is that regardless of the specific way a process has memory, like whether it's a fractional Brownian motion or something else, the probability of a flip—which is basically moving from a maximum to a minimum and back—scales universally as one over n. That means it scales as one over n, no matter what's going on underneath <ref:2506.15642#pg1>.

Mira: It suggests that memory itself imposes this universal structure on how things explore space. The authors are showing that this flip probability scales as one over n independently of the underlying stochastic process, which is quite a strong statement given how complex those processes can be <ref:2506.15642#pg1>.

Lev: From an engineering standpoint, if you were building a system trying to explore a space, knowing that this flip probability scales as one over n gives us a very simple rule of thumb for how often we expect reversals <ref:2506.15642#pg1>. It suggests that the specific microscopic details might matter more in determining the prefactor than the fundamental scaling law itself.

Kai: Right, so it's not just about whether you have memory or not; it’s about what happens when you *do* have memory and how that manifests in the exploration pattern. The paper sets up this framework by classifying different types of memory into three broad classes thirty-six <ref:2506.15642#pg3,into three broad classes 36>.

Mira: That classification is key because the authors show that for each class, there's a specific prefactor that tells us more about the system, like how it's behaving or what kind of dynamics it has. For instance, in class one, which is where flips are independent of each other asymptotically, the prefactor A is just phi <ref:2506.15642#pg3>.

Lev: If we were trying to run this on real hardware, knowing that the prefactor changes based on the memory class means we'd need a way to experimentally identify which class we're in first. You can't just measure the scaling; you have to know what mechanism is causing it before you can predict exactly how often those flips will happen.

Paper summary: Kai: Exactly, and that leads into the real-world validation they do. They test this universal one over n scaling law on several non-Markovian models, including Lévy walks and fractional Brownian motion, and they also look at financial data like the SPX index six <ref:2506.15642#pg1>.

Mira: What's interesting is that even when they apply this to real datasets from biological tracers or DNA sequences, the universal one over n scaling holds <ref:2506.15642#pg1>. This implies that this relationship between memory and exploration isn't just a feature of abstract mathematics; it shows up in complex systems we see everywhere <ref:2506.15642#pg1>.

Lev: But what about the caveats? The paper mentions some lower bounds, like for marginal random walks, where the flip probability is at least proportional to one over n times some other term that scales with a log of n and another log of log n Appendix I. That means there are specific regimes where we can get a better estimate than just the universal scaling.

Kai: That’s a good point because it shows that the universal scaling is an asymptotic limit, and for smaller systems or in specific walk regimes, those other terms matter a lot. It's not always just one over n times some constant <ref:2506.15642#pg1>.

Mira: The paper also discusses how to compute this flip probability on actual data using a windowing method Appendix C. That’s the practical side—how do we actually get these numbers from time series when we don't have perfect knowledge of the underlying process?

Lev: And they even tackle the more complex class three processes, like the saturating self-interacting random walk. They use a framework called Brownian web construction to find that exact prefactor for class three, which is four over pi squared Appendix G <ref:2506.15642#pg1>. That sounds like a lot of machinery to get one specific constant down.

Kai: The implication for someone listening just tuning in is that we don't need to know the exact physics of a market or a biological system to predict when it might reverse its direction of exploration; we just need to know how many sites have been visited.

Mira: So the paper suggests that memory-driven processes share a fundamental statistical structure regarding reversals, and this structure is simple enough to be universal across different physical models. The authors conclude that flips are a fundamental statistical feature that transcends the microscopic details of the underlying dynamics <ref:2506.15642#pg1>.

Paper summary: Lev: And they also point out that conditioning on how many distinct sites have been visited actually induces universal behavior in these memory-driven processes, which is pretty powerful for setting up models. That's what we can use to constrain the possibilities before we even start simulating Appendix I.

Kai: So, to sum up this paper "Flips Reveal the Universal Impact of Memory on Random Explorations", it establishes a simple one over n scaling law for flips in one dimension that holds across three different classes of memory-driven processes <ref:2506.15642#pg1,Flips Reveal the Universal Impact of Memory on Random Explorations>. It shows that while the specific dynamics matter for the prefactor, the fundamental scaling is universal.

Mira: And it gives us a tool to probe memory effects in real-world systems, like financial indices or biological sequences, by looking at how often exploration patterns reverse direction. This framework is robust because it doesn't rely on knowing every single microscopic detail of the process <ref:2506.15642#pg1>.

Lev: The limitation they point out is that the method relies on the assumption that we can identify and classify the memory class beforehand, and for some cases, like marginal random walks, you still have to deal with those other lower bounds Appendix I.

Kai: It’s a useful metric because it gives us a simple prediction about reversal events without needing to solve the entire complex dynamics of the system. It’s a way to filter out the noise and focus on this fundamental statistical property.

Mira: And that prefactor A, which depends on whether you're in class one, two, or three, acts as a diagnostic tool to tell us what kind of memory is dominating the exploration process in that specific system <ref:2506.15642#pg3>.

Lev: It’s interesting how this connects to the way we think about error correction; if you can characterize the memory class, you might be able to design better error correction schemes for those explorations.

Kai: Yeah, so this paper gives us a clean statistical feature that we can measure and predict across many different fields. It suggests that even in messy, complex environments, there's an underlying simple rule governing how things explore their space <ref:2506.15642#pg1>.

Conclusion: Kai: So, we're wrapping up this look at "Flips Reveal the Universal Impact of Memory on Random Explorations." The authors are showing that in one dimension, when you look at random exploration, the way a process flips direction—moving from a maximum to a minimum—is governed by a very simple rule.

Mira: It turns out that whether it's fractional Brownian motion or something else with long-range memory, the scaling law for those flips is exactly one over n. That means the number of sites visited determines how often you expect those reversals to happen, regardless of the specific physics driving the movement.

Lev: From an engineering standpoint, that’s a relief because it means we don't need to model every single microscopic detail just to predict if something will turn around. We only need to know the number of steps taken so far.

Kai: Exactly. It moves the focus from trying to understand every tiny force in a system to just measuring how much exploration has happened, which is manageable for experimental setups.

Mira: The authors found that this scaling holds even when they look at real-world data like financial indices or DNA sequences, which shows this isn't just an abstract math trick; it’s a feature of how complex things explore their space.

Lev: But there are still caveats. They pointed out that the specific prefactor, the part that tells you *how* likely it is to flip, does depend on what kind of memory you have—whether it's class one, two, or three.

Kai: Right. So while the one over n scaling is universal across classes, the specific constant in front of it gives us a fingerprint of the underlying dynamics. That prefactor is where the system-specific information lives.

Mira: It’s about getting that right kind of memory classification because that prefactor tells you which physical process you're actually dealing with. That's what makes this framework useful for real science, not just theory.

Lev: And for us building error correction hardware, knowing the class helps us decide what kind of noise we are fighting against when designing those correction codes.

Kai: So, the big picture is that memory doesn't just complicate things; it imposes a universal structure on exploration that we can measure across almost any physical system. That opens up a new way to predict reversal events in anything from markets to biology.

J. Brémont, L. Régnier, A. Barbierℾ-Chebbah, R. Voituriez, O. Bénichou

Laboratoire de Physique Théorique de la Matière Condensķee, CNRS, UPMC · Laboratoire Jean Perrin, CNRS, UPMC · Laboratoire de Physique de l’Ecole Normale SupƉerieure, CNRS, ENS & PSL University · Decision and Bayesian Computation Department CNRS UMR 3751 Institut Pasteur CNRS Paris · Epimethée, Inria Paris

cond-mat.stat-mech, physics.app-ph

Submitted: 2025-06-18

Updated: 2025-06-18

Journal ref: Nat Commun 17, 453 (2026)

DOI: 10.1038/s41467-025-67139-z

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 73/100

The gist: Flips reveal the universal impact of memory on random explorations by demonstrating that in one dimension, the probability of a flip scales as 1/n independently of the underlying stochastic process.

Key concepts

Flip
A flip is defined as a reversal in exploration, meaning the walker transitions from its current maximum visited site to its current minimum visited site, or vice-versa. It serves as a direct measure of persistence and memory in the exploration process.
Universal Scaling Law (1/n)
The central result is that for one-dimensional random explorations, the probability of a flip scales universally as 1/n. This scaling is independent of the specific type of underlying stochastic process, demonstrating a fundamental statistical feature arising from memory effects.
Memory Classes
The paper categorizes different types of memory in scale-invariant processes into three classes: independent Random Walks (RWs), Fractional Brownian Motion (fBM) with long-range correlations, and dynamically generated memory from self-repelling RWs. Each class has a specific prefactor that characterizes its unique memory behavior.

Terminology

Summary

Flips reveal the universal impact of memory on random explorations by demonstrating that in one dimension, the probability of a flip scales as 1/n independently of the underlying stochastic process. p. 0

Introduction to Flips and Memory Effects

The concept of a flip is introduced to quantify reversals in exploration, defined as transitions from a maximum to a minimum, or vice-versa, which serves as a direct measure of persistence in exploration. While this reduces to the classical splitting probability in Markovian systems, the authors show that the flip probability follows a strikingly simple and universal law: it decays inversely with the number of sites visited, as 1/n, independently of the underlying stochastic process. p. 0 This universality arises because flips occur when traversing from one edge of its own dynamically explored region to the other, introducing memory and aging into the observable itself. p. 0

Derivation of Universal Scaling Law

The central result is a remarkably simple and universal scaling law: in one dimension, the probability of a flip scales as πn ∝ 1/n. This contrasts sharply with the classical splitting probability qn, for which the decay explicitly depends on ϕ. p. 1 The scaling argument leading to the 1/n decay shows that α = 1, and thus the general behavior πn ∼ A/n. p. 1 The prefactor A remains process-specific, serving as a fingerprint of the underlying microscopic dynamics. p. 1

Classification of Memory Classes and Pre-factors

The authors classify memory in scale-invariant processes into three broad classes [36]: (I) RWs for which flips are independent, (II) fractional Brownian Motion (fBM) characterized by long-range temporal correlations, and (III) dynamically generated memory from self-repelling RWs. The full asymptotics of πn is given by πn ∼ A/n, where A depends on the class:

  1. Class (I): A = ϕ. p. 2

  2. Class (II): A = 1 − Cε, where C is a specific constant. p. 2

  3. Class (III): A = 4/π2. p. 2

Validation and Real-World Applications

The theoretical prediction of the universal 1/n scaling law is validated on paradigmatic non-Markovian models across classes (I)-(III) through numerical simulations. Experimental relevance is demonstrated by analyzing effectively one-dimensional datasets from diverse origins, including biological tracers, DNA sequences, and financial indices. p. 2 The universal 1/n scaling law holds even when the specific mechanisms responsible for non-Markovian behavior are unknown. p. 2

Conclusion and Broader Implications

The findings overturn conventional expectations by showing that memory gives rise to a universal structure governing the dynamics of exploration, suggesting that flips constitute a fundamental statistical feature. Conditioning on the number of distinct sites visited induces universal behavior in memory-driven processes. p. 4 The prefactor A encodes system-specific memory effects and quantifies the system’s propensity to reverse its direction of exploration. p. 3 Flips offer a simple and model-free tool for probing memory-driven exploration in real-world systems. p. 3

Methods for Computing Flip Probabilities

The authors employ a windowing method to address the issue of ensemble averaging from single trajectories. For real datasets, they separate time series into overlapping sub-trajectories and define flips based on whether the time series exceeds its current range. p. 9 The analysis extends to higher dimensions, showing that the universal scaling law holds for compact, symmetric and asymptotically scale-invariant RWs. p. 4 In the case of class (III), the exact prefactor A is determined by computing path integrals involving local time profiles using the Brownian web framework. p. 2 The calculation yields A = 4/π2. p. 2

Scaling in Marginal and Transient Cases

For marginal RWs, a lower bound for the flip probability is established as πn ≳ p0n × p1n, where p1n scales as 1/√log n log log n. For transient RWs, a lower bound is πtransient n ≳ p0n × p1n ∝ 1/nµ. p. 4 These bounds capture specific feasible mechanisms for a flip to occur in these regimes. p. 2 p. 2 The alternative definition of flips, based on transitioning from a newly visited site to the origin, also yields a scaling of 1/n. p. 5

Model Descriptions and Parameters

The paper describes several paradigmatic non-Markovian processes used for validation: L´evy walks are jump processes with step lengths drawn from a power-law distribution p(l) ∝ l − β−1. Fractional Brownian Motion (fBM) is a Gaussian process with stationary increments defined by the Hurst exponent H. p. 5 The Saturating Self-Interacting RW (SATW) involves transition probabilities depending explicitly on the visitation history. p. 5 The Run-and-Tumble particle is an active particle model characterized by velocity changes at exponentially distributed times. p. 8

Data Analysis and Results

The analysis of financial indices SPX, N225, and Dow-Jones shows that flips correspond to market drawdown events or corrections. Remarkably, all three financial indices exhibit a similar prefactor A ≈ 0. p. 38, pointing to a common underlying mechanism across markets. p. 3 The universal scaling law is confirmed numerically for various real-world datasets. p. 4 The analysis of log-indices for financial data confirms the robustness of the framework even in weakly biased systems. p. 5 p. 5

Summary of Key Findings

The paper concludes that flips constitute a fundamental statistical feature that transcends microscopic details, leading to two central insights. First, flips are a fundamental statistical feature that transcends microscopic details of the underlying dynamics. p. 4 Second, conditioning on the number of distinct sites visited induces universal behavior in memory-driven processes. p. 4 The universal 1/n scaling law is robust across a wide range of stochastic processes and real-world systems. p. 4 The prefactor A provides a valuable diagnostic tool for distinguishing memory classes. p. 2

References

[1] A. Dussutour, V. p. 6 Fourcassi´e, D. p. 6 Helbing, and J. p. 6-L. Deneubourg, Nature 428, 70 (2004)

[2] L. Giuggioli, J. p. 6 R. Potts, and S. p. 6 Harris, PLOS Computational Biology 7, 1 (2011)

[3] O. B´enichou, C. p. 6 Chevalier, J. p. 6 Klafter, B. p. 6 Meyer, and R. p. 6 Voituriez, Nature Chemistry 2, 472 (2010)

[4] Decision and Bayesian Computation, USR 3756 (C3BI/DBC) & Neuroscience department CNRS UMR 3751, Institut Pasteur, CNRS, Paris, France

[5] Epimeth´ee, Inria, Paris, France

[6] J.-P. Bouchaud and M. p. 6 Potters, Theory of Financial Risk and Derivative Pricing: From Statistical Physics to Risk Management, 2nd ed. p. 6 (Cambridge University Press, Cambridge, 2003)

[7] L. R. Goldberg and O. p. 6 Mahmoud, Mathematics and Financial Economics 11, 275 (2017)

[8] J. Hull, Options, futures, and other derivatives, 6th ed. p. 6 (Pearson Prentice Hall, Upper Saddle River, NJ [u. p. 6a.], 2006)

[9] J. Rudnick and G. p. 6 Gaspari, Science 237, 384 (1987)

[10] L. R´egnier, M. p. 0 Dolgushev, S. p. 6 Redner, and O. p. 6 B´enichou, Nature Communications 14, 618 (2023)

[11] H. Larralde, P. p. 6 Trunfio, S. p. 6 Havlin, H. E. Stanley, and G. p. 6 H. Weiss, Nature 355, 423 (1992)

[12] B. D Hughes, Random Walks and Random Environments: Random Walks (Clarendon Press, 1995)

[13] V. E Tarasov, Fractional Dynamics: Applications of Fractional Calculus to Dynamics of Particles, Fields and Media, edited by A. p. 6 C. J. Luo and N. H. Ibragimov, Nonlinear Physical Science, Vol. p. 6 0 (Springer, Berlin, Heidelberg, 2010)

[14] J M Phillip et al., Communications biology 4, 81 (2021)

[15] B. B Mandelbrot and J. p. 6 W. Van Ness, SIAM Review 10, 422 (1968)

[16] H Flyvbjerg, Nature Physics 17, 771 (2021), publisher: Nature Publishing Group

[17] L Tröger et al., Proceedings of the National Academy of Sciences 121, e2312611124 (2024)

[18] A J Bray, S. N Majumdar, and G Schehr, Adv. p. 6 Phys. 62, 225 (2013)

[19] S N Majumdar et al., Physical Review Letters 104, 020602 (2010)

[23] B D Hughes, Random Walks and Random Environments: Random Walks (Clarendon Press, 1995)

[24] T Rolski et al., Stochastic Processes for Insurance & Finance (John Wiley & Sons, Ltd, 1999)

[25] D Boyer et al., Proceedings of the Royal Society B: Biological Sciences 273, 1743 (2006)

[26] P L Krapivsky et al., A Kinetic View of Statistical Physics (Cambridge University Press, Cambridge, 2010)

[27] M J Saxton et al., Biophysical Journal 66, 394 (1994)

[28] E Barkai, E Aghion, and D A Kessler, Physical Review X 4, 021036 (2014)

[29] D Boyer and C Solis-Salas, Physical Review Letters 112, 240601 (2014)

[30] B Toth, Ann Probab 23, 1523 (1995)

[31] T Baravi et al., Physical Review Letters 134, 127101 (2025)

[32] J-P Bouchaud and A Georges, Phys. Rep. 195, 127–293 (1990)

[34] We write An ∼ Bn if An/Bn → 1 as n → ∞

[35] More precisely, our analysis shows that Fn ∝ ln n, meaning the number of sites visited at the kth flip is exponential with k

[36] L R´egnier et al., Nat. Commun. 14, 6288 (2019)

[37] V Zaburdaev et al., Reviews of Modern Physics 87, 483 (2015)

[38] G Ariel et al., A Rabani et al., S Benisty, J D Partridge, R M Harshey, and A Be’er, Nature Communications 6, 8396 (2015)

[39] B Meerson, Physical Review E 107, 064122 (2023)

[40] T W Burkhardt, Journal of Statistical Mechanics: Theory and Experiment 2007, P07004 (23)

[41] D J Amit et al., Phys. Rev. B 27, 1635 (1983)

[42] A Barbier-Chebbah, O Benichou, and R Voituriez, Phys. Rev. E 102, 062115 (2020)

[43] A Barbier-Chebbah, O Benichou, and R Voituriez, Phys. Rev. E 102, 062115 (ref:7)

[46] For fBM (class II), we use A = e − Cε instead of A = 1 − Cε

[47] L R´egnier et al., Nat. Commun. 14, 618 (2023)

[48] See (3) for the arrival convention

[49] L R´egnier et al., Nat. Commun. 14, 618 (2023)

[50] J Br´emont et al., Physical Review Letters 134, 197103 (2025)

[52] G Molchan, Communications In Mathematical Physics 205, 97 (1999)

[53] DNA sequences are Pyrimidines/Purines HUMBMYH7 and HUMTCRADCV (1994)

[54] M J McLean et al., Journal of Molecular Evolution 47, 691 (1998)

[55] A Nec¸sulea and J R Lobry, Molecular Biology and Evolution 24, 2169 (2007)

[56] A Rej, P Seager, and J-P Bouchaud, Wilmott 2018, 56 (2018)

[57] Note that after the logarithmic transform explained in the Methods section, n cannot be interpreted as the size of the visited interval

[58] A Van Buskirk, Volatility Skew, Earnings Announcements, and the Predictability of Crashes (2011), Social Science Research Network:1740513

[59] A Bancaud et al., Nucleic Acids Research 40, 8783 (2012)

[60] A Bancaud et al., EMBO J 28, 3785 (2009/12/16/print)

[61] J Klafter and I M Sokolov, First Steps in Random Walks: From Tools to Applications (OUP Oxford, 2011)

[62] A Vezzani and R Burioni, Physical Review Letters 132, 187101 (2024)

[63] I Rhee et al., IEEE/ACM Transactions on Networking 19, 630 (2011)

[65] K Eftaxias et al., Natural Hazards and Earth System Sciences 8, 657 (2008)

[66] J Gatheral et al., Quantitative Finance 18, 933 (2018)

[67] B Hokmabad et al., Proceedings of the National Academy of Sciences 119, e2122269 (2024)

[68] J d’Alessandro et al., Nature Communications 12 (2021)

[69] A C Maggs, Europhysics Letters 147, 21001 (23)

[70] B Toth and W Werner, Probability Theory and Related Fields 111, 375 (1998)

Appendix Details

Appendix A describes the non-Markovian models used, including Fractional Brownian motion (fBm), RAP, L´evy Walk, TSAW, and PSRW. Appendix C details the computation of flip probability in real datasets using a windowing method. p. 9 Appendix G provides a rigorous derivation for the prefactor A for class (III) processes using the Brownian web framework. p. 2 The analysis of marginal and transient RWs yields lower bounds for πn, such as πmarginal n ≳ 1/n log n log log n. p. 4 The universal scaling law is confirmed numerically in Fig. p. 4 5 for various compact and non-compact RWs. p. 4

Appendix I Details

The analysis of marginal RWs establishes a lower bound for the flip probability as πmarginal n ≳ p0n × p1n. For transient RWs, the lower bound is πtransient n ≳ p0n × p1n ∝ 1/nµ. p. 4 These bounds capture specific feasible mechanisms for a flip to occur in these regimes. The analysis of log-indices for financial data confirms the robustness of the framework even in weakly biased systems. The study shows that flips reveal, and may act as a statistical signature of, an emergent structure in the statistical patterns of market dynamics. p. 3 p. 3 The universal 1/n scaling law does not reflect specific underlying mechanisms but instead offers a robust and predictive metric for reversal events. p. 3 p. 3

Appendix H Details

The numerical verification confirms that the alternative definition of flips also leads to a flip probability that decays as 1/n. The universal scaling is predicted by the physical argument presented in Eq. p. 1 (3) of the main text. p. 8 The prefactor A can now be computed, yielding A = 1 − 48 log(G) − 4γ − 1/328 log(2) ε + O(ε 2), where γ is Euler’s constant. p. 2 The study shows that flips reveal, and may act as a statistical signature of, an emergent structure in the statistical patterns of market dynamics. The universal 1/n scaling law does not reflect specific underlying mechanisms but instead offers a robust and predictive metric for reversal events.

Appendix G Details

The Brownian web provides a concrete and tractable framework to study aging in the TSAW. This framework naturally encodes the full history of the walker’s past visits, making it especially well-suited to capturing long-range temporal correlations and memory effects. The prefactor A is fully universal across all class (III) processes. p. 1 The Brownian web constructions for other processes of class (III), such as the PSRW, involve powers of Brownian motions, rather than standard Brownian paths. p. 8 This argument justifies our focus on the TSAW case. p. 1 The exact prefactor A is determined by computing path integrals involving local time profiles using the Brownian web framework. p. 3 The exact prefactor A is 4/π2. p. 3

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

  1. Universal flip statistics for random walks allows AI systems to quantify memory effects across diverse stochastic processes without requiring model-specific assumptions. This enables AI to analyze real-world data from biological tracers, DNA sequences, and financial markets using a single scaling law of 1/n.

  2. The ability to distinguish between the universal 1/n scaling of flips and process-dependent splitting probabilities (qn proportional to n-phi) allows AI systems to identify the underlying memory class (I, II, or III) of a system simply by analyzing the prefactor A.

  3. AI systems can use the prefactor A as a diagnostic tool to distinguish between different classes of non-Markovian dynamics, such as L´evy walks and saturating self-interacting random walks (TSAW), which share walk dimensions but have different scaling factors for flips.

  4. The framework extends to higher dimensional and fractal domains, enabling AI to analyze complex transport in environments like critical percolation clusters by predicting the universal 1/n scaling of flips in those settings as well.

  5. AI systems can extract a model-free indicator of genomic asymmetry by analyzing DNA sequences, using flips (transitions between G-rich and C-rich regions) as a measure, which is independent of the specific underlying sequence structure.

Sources

Related papers