Zeroes and Extrema of Functions via Random Measures
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.
Tom: We're looking at the paper: "Zeroes and Extrema of Functions via Random Measures".
Jane: Here's the gist of the paper. Zeroes and Extrema of Functions via Random Measures presents methods that provide all zeroes and extrema of a function without requiring differentiation using point process theory.
Tom: The proposed Point Process Zeroes (PPZ) method offers an intuitively appealing approach that simultaneously provides all unknown zeroes and their number, as well as the unknown points of extrema and the function maximum/minimum value over a bounded window of observation.
Jane: Theoretical Framework The mathematical formulation begins by considering a general function space F Mp M where f is continuous in (M, B(M)) and equipped with an inner product ρ to define a norm and distance.
Tom: Since the set of zeroes Ξf can be uncountable, it is treated as the realization of a random counting measure N, which defines a point process where N(W) = X x∈N I(x ∈ W).
Title and authors: Tom: So we're looking at this paper called "Zeroes and Extrema of Functions via Random Measures". It sounds like they're trying to find all the places where a function equals zero, or where it hits a peak or a valley, without having to use any kind of calculus.
Jane: That’s what it claims. They say using point process theory, they can describe not just where these zeroes are, but also how many there are and even their distribution over some window of observation. It sounds like they’re tackling that hard problem of finding those exact points without needing to differentiate the function itself.
Lu: The core idea is to treat the set of zeroes as a random set, which requires estimation or approximation because you don't know where they are or how many there are yet. They start by setting up a general function space, and then they turn that into something that can be described by a point process.
Meng: But finding those points without differentiation sounds like it might be really hard to actually implement when we're dealing with real-world data, especially if we need to deploy this at scale. What are the practical constraints here?
Tom: Well, the paper sets up an Inhomogeneous Poisson Point Process, which they call Nf. They use an intensity function that depends on the function itself, specifically using exp of minus K times the absolute value of f at a point multiplied by some Q value.
Jane: So, they model the set of zeroes using this process where the intensity is tied directly to how big or small the function's value is at that spot. They then say that as K gets really big, the expected number of zeroes in any window actually converges to the real number of zeroes there.
Lu: They suggest a way to find extrema by relating them back to the anti-derivative of the function, which is connected to how you deal with vector valued functions in a multidimensional space. For those, they approximate the derivative using finite differences and apply this point process idea again.
Meng: Vector valued functions sound much more complex when you're trying to figure out where the gradient equals zero, right? How does this method handle that complexity at scale?
Tom: The paper suggests that for multivariate functions, you can use the norm of the function, like the absolute value of all its components summed up. This allows them to apply the same logic to find zeroes in higher dimensions.
Title and authors: Jane: That’s helpful because it means we can use this approach for things that aren't just simple single numbers but vectors of values across multiple dimensions. They mention that they can also use this idea for optimization problems, like finding maximum likelihood estimates for models.
Lu: The authors are pretty ambitious here, suggesting the Adaptive Poisson Point Process method allows you to get as close to a zero or an extremum as you want by refining the search iteratively. This adaptive version is key because it avoids getting stuck in local areas where the approximation isn't good enough.
Meng: Iterative refinement sounds like it could be computationally expensive, especially if you need very high precision, say below two to the k tolerance level they mention. What about making this run on a production system?
Tom: The paper discusses using constants K and Q to control how much we reject points during the simulation. And they have an adaptive version, the APPPZ, which refines its search by building smaller windows around a potential zero as it goes.
Jane: And they point out that choosing Q is important; they recommend a value of Q equals zero point five for real functions and two for complex ones to control the rejection rates effectively.
Lu: They also applied this to complex functions, like the Riemann zeta function, which is interesting because it lets them verify known zeroes on specific strips in the complex plane. But they do flag that approximating function values for those complex functions can take a lot of time.
Meng: So, if we're dealing with something computationally intensive like that zeta function, we have to worry about the runtime before we even worry about deployment costs.
Tom: And another thing they mention is the Cox point process approach. Instead of assuming a fixed intensity function, you can treat K as a random variable. This lets you model a random intensity, which means it models conditional independence instead of strict Poisson randomness.
Jane: That moves away from the simple independent assumption and makes it more realistic for certain complex systems where the underlying rate of events isn't perfectly constant or predictable everywhere.
Lu: The authors conclude that this APPPZ method gives you all zeroes using just one realization, which is a big deal because it avoids the convergence issues you often see with iterative methods like simulated annealing. It seems to provide a complete picture in a single go.
Title and authors: Meng: If it truly finds *all* zeroes in one run without getting stuck, that’s something I’d need to see on real traffic data before I’d even think about deploying anything at scale. What happens when the function is really messy?
Tom: The paper shows it scales easily to higher dimensions for multivariate real functions, and they say all zeroes are recovered well regardless of the specific dimension p. It seems robust in that regard.
Jane: So, to wrap up on this paper, "Zeroes and Extrema of Functions via Random Measures", they present a method using point process theory to find function zeroes and extrema without needing differentiation. They show how it handles both real and complex functions, with an adaptive version that refines the search for better accuracy.
Lu: The main implication is that you can get a complete set of zeroes and their locations in one go, which is really useful for analyzing function behavior or solving equations where you need the whole picture.
Meng: For production use, the biggest hurdle seems to be controlling the computational time when dealing with complex functions, and making sure the adaptive process doesn't take too long for high precision targets.
Tom: And another thing they noted is that they can model a random intensity using K as a random variable, which is useful if you need to account for uncertainty in the underlying point process rate itself.
Jane: So, this paper gives us a powerful tool to look at functions and see where they are zero or extreme without relying on traditional calculus. It opens up new ways to analyze these kinds of problems across different mathematical settings.
Lu: The method seems most promising for situations where the set of zeroes is uncountable, treating it as a realization of a random counting measure, which is a very general way to model that kind of randomness.
Meng: I'm just thinking about how much tuning we’d need for those parameters K and Q to make sure it actually works reliably on something messy. If the settings are too sensitive, the whole thing breaks.
Tom: It sounds like they gave us a solid framework, but now we need to see if they can tame that complexity when we plug in real-world noisy data instead of clean mathematical functions.
Jane: That’s exactly what the next steps should look like—testing how this APPPZ handles the noise and uncertainty we actually see in the real world.
The paper's summary: Tom: So this paper is essentially saying we can find all the places where a function hits zero or its maximums and minimums without ever using calculus, just by treating those points as a random set.
Jane: That’s right, Tom, it uses point process theory to describe the zeroes and extrema of a function. But what does that actually mean for us? Does it just mean we skip the math?
Lu: Not at all. It means they treat the set of zeroes like a physical realization of a random counting measure. This is how they get past the problem where finding those exact points is usually impossible because you don't know where they are beforehand.
Meng: So, if I’m trying to ship a model and I need to know all the local minima or maxima, this approach suggests we can find them simultaneously with all the zeroes in one go? That sounds like a huge win for debugging.
Tom: Exactly. Because it uses this point process idea, you get not just where the function is zero, but also how many there are and where those peaks and valleys are over some window of observation. It’s like getting a full map instead of just a few known landmarks.
Jane: That sounds really intuitive for someone who doesn't read math papers; it’s about getting the whole picture at once. But they have to deal with complexity, right? They mention multivariate functions where you have more than one input variable.
Lu: They do, and that’s where the method gets interesting because it scales well to higher dimensions by using something called the norm of the function. This lets you apply the same zero-finding logic even when you're dealing with a vector of values instead of just a single number.
Meng: Scaling up sounds good for theory, but what about running this on actual production data? If I’m looking at a high-dimensional problem, does it still keep the computational load manageable? Or is it just another slow simulation?
Tom: The authors point to an adaptive version they call the APPPZ. It lets the AI refine its search iteratively to get down to a very small tolerance level, like two to the k level they mention. It avoids getting stuck in those local traps that iterative methods usually fall into.
Jane: So instead of running something for forever hoping it gets close, this method actively refines where it’s looking next based on what it finds so far. That sounds much more efficient than just brute force searching the whole space.
Lu: And they also look at modeling uncertainty by treating a parameter called K as a random variable, which turns the process into a Cox point process. This lets you model that randomness in the underlying rate of events, which is pretty powerful for real-world scenarios where things aren't perfectly predictable.
Meng: Modeling uncertainty is important for deployment because I can’t just assume everything is perfectly Poisson distributed everywhere on my data. If the intensity itself varies randomly, that changes how much noise I have to filter out in the final model output.
Tom: Right, so they are giving us a way to handle that messy reality—the randomness in both where the zeroes are and how often they occur. It moves us away from assuming perfect independence for every single point.
Jane: So, the big implication here is that we get a complete set of information—zeroes and extrema—from just one run, which cuts down on those convergence headaches you see with other methods like simulated annealing.
Lu: Yeah, they really want to provide the entire set of zeroes using just one realization. That’s a significant step because it’s much more robust than relying on iterative algorithms that might get stuck in a local characteristic of the target function.
Meng: If this works as well as they claim across different dimensions and functions, then maybe we can start thinking about applying this kind of zero-finding logic to optimization problems, like finding those maximum likelihood estimates for my models.
Tom: That’s the other big thing they touch on—applying it to optimization problems using the gradient function. It shows that this isn't just theoretical math; it’s a tool that can actually help solve practical engineering and data science problems.
Jane: So, in short, this paper gives us a way to see the entire structure of a function’s behavior—all its zeroes and all its turning points—without having to rely on traditional derivative calculations. We need to see how we can translate that framework into something we can actually run reliably on real data.
The paper's improvements: Tom: So we’re looking at how they improved their initial idea for finding zeroes and extrema in this function space paper, and it’s actually pretty interesting because they fixed some of those convergence issues we talked about before.
Jane: Right, so what were the main problems with the first version? I remember them saying iterative methods could get stuck in local characteristics of the target function.
Lu: That’s right, and they addressed that by using a single realization instead of relying on a sequence of guesses. This means they aim to provide all zeroes using just one realization, which avoids those convergence issues inherent in iterative methods like simulated annealing.
Meng: Avoiding the convergence problem is good for theory, but from an engineering standpoint, what does that mean for deployment? If it only works with one run to get the whole picture, that’s a big assumption when you need reliability under constant load.
Tom: It means they are trying to offer a complete set of zeroes and their number in one go. They want to avoid the scenario where an iterative method might fail before it even gets close enough to be useful.
Jane: So, if we think about this for our AI systems, it suggests that instead of running a process until you hit a stopping rule, you’re looking at a single snapshot and getting all the answers from that one snapshot. That sounds much more deterministic.
Lu: And they also introduced an adaptive version called APPPZ. This algorithm refines the search by iteratively reducing the tolerance level, which helps achieve a desired end tolerance T end, even if it takes a lot of time for very low precision targets, like two to the k level.
Meng: How much time are we talking about? If I need that kind of precision and it takes forever to run on a standard GPU cluster, then the benefit is limited unless the function is simple.
Tom: The authors admit that for certain functions and specific choices of parameters K and Q, achieving that very low tolerance can take a lot of time. But they showed it scales well to higher dimensions, which is good because it means we can use this logic for more complex multivariate vector valued functions.
Jane: So they’re saying the method is robust enough for high-dimensional problems even if the search takes longer to nail down that fine detail. It’s not just a trick for simple 1D functions anymore <ref:2511.10293#pg1>.
Lu: Exactly, and it also applies to optimization problems, like finding maximum likelihood estimates for multivariate normal models, by applying this point process idea to the gradient function itself. That broadens the scope beyond just finding zeroes of a single function.
Meng: If that means we can use this for optimization where we’re trying to find the best settings for a model, it could be useful before I even worry about scaling up to production traffic.
Tom: It’s about showing that this is a general way to tackle these kinds of problems across different mathematical settings. It shows the framework has some real potential beyond just solving one specific equation.
Jane: So, the main thing here is that they’re giving us a more complete picture—zeroes, extrema, and their counts—in a single run while also showing ways to make it adaptive for high precision. But we still have to figure out how to tame the computational time for those fine tolerances.
Lu: And as we move forward, they are thinking about treating K as a random variable to model a Cox point process, which lets us handle the inherent randomness in the rate of events itself. That’s their next big direction for more realistic modeling.
Meng: Modeling that random intensity is definitely something I’m interested in; it means we can account for uncertainty in the underlying process rather than assuming perfect control over everything.
Conclusion: Tom: So to wrap this up, we’re looking at the paper "Zeroes and Extrema of Functions via Random Measures," which basically gives us a method to find all zeroes and extrema without using any calculus.
Jane: That’s right, Tom, it’s about using point process theory to describe the locations of zeroes and turning points in a function. It’s like getting a complete map of where the function is behaving weirdly or hitting zero.
Lu: The main idea is that we can treat the set of zeroes as a random set, which lets us find them all in one go without getting stuck on local characteristics of the target function.
Meng: So, if this works across different dimensions for multivariate functions and optimization problems, it means we might be able to apply this kind of zero-finding logic to things like finding the best settings for a machine learning model.
Tom: It’s about showing that this is a general way to tackle these kinds of problems across different mathematical settings, which is pretty cool because it opens up new avenues for solving equations.
Jane: We still have to worry about the practical side, though. The APPPZ method needs time to refine the search for high precision targets, and controlling those parameters K and Q is crucial for making sure it actually works reliably in a real-world scenario.
Lu: Exactly; we’re talking about managing that computational time versus getting that fine detail down to a very small tolerance level, like two to the k level.
Meng: I still think we need to see how this runs on actual traffic data before I can even think about deploying anything, especially if the search time is too long for my production environment.
Tom: It’s about showing that this is a tool that can actually help solve practical problems, not just something purely theoretical. The fact that it handles multivariate functions suggests good scalability.
Jane: So, we’re looking at a powerful tool to analyze function behavior—where the zeroes and extrema are—without needing traditional derivative calculations. But we still have to figure out how to tame the computational time for those fine tolerances and test it on noisy data.
Lu: And future work they mentioned involves treating K as a random variable, which lets us model a Cox point process instead of assuming a fixed intensity function. That’s where the real modeling ambition is going next.
Meng: Modeling that random intensity is definitely something I’m interested; it means we can account for uncertainty in the underlying rate of events rather than assuming perfect control over everything.
Tom: So, yeah, "Zeroes and Extrema of Functions via Random Measures" gives us a way to see the whole structure at once using point process theory. It’s a solid framework to build on.
Jane: Yeah, it’s a lot of math for what it does—giving you all the zeroes and extrema in one go. It definitely changes how we think about function analysis, even if we still have those practical hurdles to overcome with time and noise.
University of Missouri
stat.ME, math.OC, stat.CO
Submitted: 2025-11-13
Updated: 2026-11-11
Comments: Corrected the dominating measure to have both atomic and nonatomic parts. Fixed typos. Function extrema and zeroes; Optimization; Poisson point process; Random counting measures; Riemann's zeta function
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 70/100
The gist: Zeroes and Extrema of Functions via Random Measures presents methods that provide all zeroes and extrema of a function without requiring differentiation using point process theory.
Key concepts
- Point Process Zeroes (PPZ)
- This is the core method that treats the set of function zeroes as a random point process. Instead of solving equations iteratively, it models these zeroes using probability theory to find all locations and count them at once in a given observation window.
- Inhomogeneous Poisson Point Process (IPPP)
- The paper uses an IPPP to model the distribution of zeroes. The intensity function dictates where zeroes are expected to occur; by setting this intensity based on the function's value, the process helps estimate how many zeroes exist in a specific region.
- Adaptive PPPZ (APPPZ)
- This is a refinement algorithm that improves the accuracy of finding zeroes and extrema. It iteratively shrinks the search area around potential points, allowing AI systems to achieve very high precision (low tolerance) by focusing their search where the function behavior is most complex.
Terminology
Summary
Zeroes and Extrema of Functions via Random Measures presents methods that provide all zeroes and extrema of a function without requiring differentiation using point process theory. p. 0 The proposed Point Process Zeroes (PPZ) method offers an intuitively appealing approach that simultaneously provides all unknown zeroes and their number, as well as the unknown points of extrema and the function maximum/minimum value over a bounded window of observation. p. 1
Theoretical Framework
The mathematical formulation begins by considering a general function space F Mp M where f is continuous in (M, B(M)) and equipped with an inner product ρ to define a norm and distance. Since the set of zeroes Ξf can be uncountable, it is treated as the realization of a random counting measure N, which defines a point process where N(W) = X x∈N I(x ∈ W). p. 3 The core idea is to exploit this representation to find zeroes or extrema of f. p. 2
Zeroes via Poisson Point Processes
The paper models the set of zeroes using an Inhomogeneous Poisson Point Process (IPPP) Nf with intensity function λK(x) = exp −Kf(x)Q. The expected number of zeroes over a window W is given by E [Nf (W)] = Z W exp −Kf(x) Q µp (dx). p. 5 As K → ∞, the expected number of zeroes converges to the actual number of zeroes in the window, E [Nf (W)] → #Ξf ∩ W. p. 5 The joint distribution of the number and locations asymptotically approaches that of a Poisson Point Process of Zeroes (PPPZ). p. 6
Identifying Extrema and Generalizations
The set of zeroes Ξf is related to the extrema of the anti-derivative g(x) = R f(x)dµ1(x), where f = g′. For multivariate functions, zeroes of the gradient ∇cf(x) are sought by approximating the derivative using finite differences. p. 7 The PPPZ method is applied to the modification of the intensity function λK(x) = exp −K∥f(x) Q. p. 7 This allows for finding zeroes of a multivariate, vector-valued function f: Mp→ Mq by considering∥f(x)∥ = X q i=1 fi(x). p. 7
Implementation and Practical Considerations
In practice, the simulation method relies on constants K > 0 and Q > 0 to control rejection rates and a tolerance variable T ol. The algorithm includes an adaptive version (APPPZ) that refines the search by iteratively reducing the tolerance. p. 2 Choosing appropriate values for Q is crucial, with recommendations being Q = 0. p. 85 for real functions and Q = 2 for complex functions. p. 0 The adaptive procedure refines the location of zeroes by building smaller windows centered at potential zeroes. p. 8
Complex Function Applications
The method is applied to complex functions, such as polynomials and Riemann’s zeta function, where the intensity function is defined using the modulus of the function. For Riemann’s zeta function, the PPPZ is used to illustrate and verify known zeroes on specific strips in the complex plane. p. 0 The computational time required for approximating function values η(s) can be a major drawback when dealing with complex functions. p. 0
Multivariate Analysis
For multivariate real functions, the method scales easily to higher dimensions, and results show that all zeroes are recovered well regardless of the specific dimension p. Optimization problems, such as maximum likelihood estimation for a multivariate normal model, can also be solved by applying PPPZ to the gradient function. p. 0 The method is capable of recovering many local minima and maxima simultaneously in challenging scenarios. p. 7
Concluding Remarks
The authors conclude that the APPPZ provides all zeroes using a single realization, avoiding the convergence issues inherent in iterative methods like simulated annealing. Future work may involve treating K as a random variable to model Cox point processes. p. 3 The method's advantage lies in providing the entire set of zeroes, which is vital for analyzing function behavior and solving equations.
Acknowledgements
The authors acknowledge Professor Stamatis Dostoglou for constructive comments and suggestions on an earlier version of the manuscript.
References
[4] O. Barndorff-Nielsen. Information and exponential families: in statistical theory. p. 0
[7] P. Borwein. An efficient algorithm for the Riemann zeta function. p. 0
[8] A. Brochard, B. B laszczyszyn, S. Zhang, and S. Mallat. Particle gradient descent model for point process generation. p. 3
[34] A C Micheas. p. 0 Cox point processes: why one realisation is not enough. p. 2
[47] W. Wu and A C Micheas. p. 0 p. 0 Modeling Fourier expansions using point processes on the complex plane with applications. p. 2 p. 2 p. 2 p. 2 p. 2 p. 2 p. 2
[50] Z. Zhou, D. S. Matteson, D. B. Woodard, S. G. Henderson, and A C Micheas. p. 0 A spatio-temporal point process model for ambulance demand. p. 3 p. 3
[35] A C Micheas. p. 0 Random mixture Cox point processes. p. 3
[48] W. Wu and A C Micheas. A new construction of covariance functions for Gaussian random fields. p. 2 p. 2 p. 2 p. 2 p. 2 p. 2 p. 2
[46] M. N. M Van Lieshout and Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions. p. 8 p. 8 p. 8 p. 8 p. 8 p. 8 p. 8
[49] C. Y Yau and J. M Loh. A generalization of the Neyman–Scott process. p. 0 p. 0 p. 0 p. 0 p. 0 p. 0
[30] F Lavancier, J Møller, and E Rubak. Determinantal point process models and statistical inference. p. 2 p. 2 p. 2 p. 2 p. 2
[44] X Tang and L Li. Multivariate temporal point process regression. p. 0 p. 0 p. 0 p. 0 p. 0 p. 0
[36] J Møller. Spatial Statistics and Computational Methods. p. 0 p. 0 p. 0 p. 0 p. 0
[37] J Møller and R P Waagepetersen. Statistical inference and simulation for spatial point processes. p. 2 p. 2 p. 2 p. 2 p. 2
[40] D A Pierre. Optimization theory with applications. p. 2 p. 2 p. 2 p. 2 p. 2
[43] E Spodarev, editor. Stochastic geometry, spatial statistics and random fields: asymptotic methods. p. 0 p. 0 p. 0 p. 0 p. 0 p. 0
[45] M N M Van Lieshout. Markov point processes and their applications. p. 2 p. 2 p. 2 p. 2 p. 2 p. 2
[38] P Pedregal. Introduction to optimization, volume 46. p. 0 p. 0 p. 0 p. 0 p. 0
[41] U Sadana, A Chenreddy, E Delage, A Forel, E Frejinger, and T Vidal. A survey of contextual optimization methods for decision-making under uncertainty. p. 9 p. 9 p. 9 p. 9 p. 9
[25] A Karr. Point processes and their statistical inference. p. 2 p. 2 p. 2 p. 2 p. 2
[42] J Sondow. Zeros of the alternating zeta function on the line r (s)= 1. p. 7
[43] E Spodarev, editor. Stochastic geometry, spatial statistics and random fields: asymptotic methods.
[44] X Tang and L Li. Multivariate temporal point process regression.
[45] M N M Van Lieshout. Markov point processes and their applications.
[46] M N M Van Lieshout. Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions.
[47] W Wu and A C Micheas. p. 0 p. 0 p. 0 p. 0 p. 0 p. 0 Modeling Fourier expansions using point processes on the complex plane with applications.
[48] W Wu and A C Micheas. p. 0 p. 0 p. 0 p. 0 p. 0 p. 0 A new construction of covariance functions for Gaussian random fields.
[49] C Y Yau and J M Loh. A generalization of the Neyman–Scott process.
[50] Z Zhou, D S Matteson, D B Woodard, S G Henderson, and A C Micheas. p. 0 A spatio-temporal point process model for ambulance demand.
[33] A C Micheas. p. 0 Hierarchical Bayesian modeling of marked non-homogeneous Poisson processes with finite mixtures and inclusion of covariate information. p. 2
[47] W Wu and A C Micheas. Modeling Fourier expansions using point processes on the complex plane with applications.
[48] W Wu and A C Micheas. A new construction of covariance functions for Gaussian random fields.
[46] M N M Van Lieshout. Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions.
[49] C Y Yau and J M Loh. A generalization of the Neyman–Scott process.
[30] F Lavancier, J Møller, and E Rubak. Determinantal point process models and statistical inference.
[44] X Tang and L Li. Multivariate temporal point process regression.
[36] J Møller. Spatial Statistics and Computational Methods.
[37] J Møller and R P Waagepetersen. Statistical inference and simulation for spatial point processes.
[40] D A Pierre. Optimization theory with applications.
[43] E Spodarev, editor. Stochastic geometry, spatial statistics and random fields: asymptotic methods.
[45] M N M Van Lieshout. Markov point processes and their applications.
[38] P Pedregal. Introduction to optimization, volume 46.
[41] U Sadana, A Chenreddy, E Delage, A Forel, E Frejinger, and T Vidal. A survey of contextual optimization methods for decision-making under uncertainty.
[25] A Karr. Point processes and their statistical inference.
[46] M N M Van Lieshout. Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions.
[47] W Wu and A C Micheas. Modeling Fourier expansions using point processes on the complex plane with applications.
[48] W Wu and A C Micheas. A new construction of covariance functions for Gaussian random fields.
[49] C Y Yau and J M Loh. A generalization of the Neyman–Scott process.
[30] F Lavancier, J Møller, and E Rubak. Determinantal point process models and statistical inference.
[44] X Tang and L Li. Multivariate temporal point process regression.
[36] J Møller. Spatial Statistics and Computational Methods.
[37] J Møller and R P Waagepetersen. Statistical inference and simulation for spatial point processes.
[40] D A Pierre. Optimization theory with applications.
[43] E Spodarev, editor. Stochastic geometry, spatial statistics and random fields: asymptotic methods.
[45] M N M Van Lieshout. Markov point processes and their applications.
[38] P Pedregal. Introduction to optimization, volume 46.
[41] U Sadana, A Chenreddy, E Delage, A Forel, E Frejinger, and T Vidal. A survey of contextual optimization methods for decision-making under uncertainty.
[25] A Karr. Point processes and their statistical inference.
[46] M N M Van Lieshout. Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions.
[47] W Wu and A C Micheas. Modeling Fourier expansions using point processes on the complex plane with applications.
[48] W Wu and A C Micheas. A new construction of covariance functions for Gaussian random fields.
[49] C Y Yau and J M Loh. A generalization of the Neyman–Scott process.
[30] F Lavancier, J Møller, and E Rubak. Determinantal point process models and statistical inference.
[44] X Tang and L Li. Multivariate temporal point process regression.
[36] J Møller. Spatial Statistics and Computational Methods.
[37] J Møller and R P Waagepetersen. Statistical inference and simulation for spatial point processes.
[40] D A Pierre. Optimization theory with applications.
[43] E Spodarev, editor. Stochastic geometry, spatial statistics and random fields: asymptotic methods.
[45] M N M Van Lieshout. Markov point processes and their applications.
[38] P Pedregal. Introduction to optimization, volume 46.
[41] U Sadana, A Chenreddy, E Delage, A Forel, E Frejinger, and T Vidal. A survey of contextual optimization methods for decision-making under uncertainty.
[25] A Karr. Point processes and their statistical inference.
[46] M N M Van Lieshout. Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions.
[47] W Wu and A C Micheas. Modeling Fourier expansions using point processes on the complex plane with applications.
[48] W Wu and A C Micheas. A new construction of covariance functions for Gaussian random fields.
[49] C Y Yau and J M Loh. A generalization of the Neyman–Scott process.
[30] F Lavancier, J Møller, and E Rubak. Determinantal point process models and statistical inference.
[44] X Tang and L Li. Multivariate temporal point process regression.
[36] J Møller. Spatial Statistics and Computational Methods.
[37] J Møller and R P Waagepetersen. Statistical inference and simulation for spatial point processes.
[40] D A Pierre. Optimization theory with applications.
[43] E Spodarev, editor. Stochastic geometry, spatial statistics and random fields: asymptotic methods.
[45] M N M Van Lieshout. Markov point processes and their applications.
[38] P Pedregal. Introduction to optimization, volume 46.
[41] U Sadana, A Chenreddy, E Delage, A Forel, E Frejinger, and T Vidal. A survey of contextual optimization methods for decision-making under uncertainty.
[25] A Karr. Point processes and their statistical inference.
[46] M N M Van Lieshout. Non-parametric adaptive bandwidth selection for kernel estimators of spatial intensity functions.
[47] W Wu and A C Micheas. Modeling Fourier expansions using point processes on the complex plane with applications.
[48] W Wu and A C Micheas. A new construction of covariance functions for Gaussian random fields.
[49] C Y Yau and J M Loh.
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
-
Point Process Zeroes (PPZ) capability allows AI systems to find all zeroes of a function and their number simultaneously over a bounded window in a single realization, which avoids iterative methods
locked into local characteristics of the target function (e.g., simulated annealing) or do not converge.
-
The method scales easily to higher dimensions by construction, enabling AI systems to apply the same zero-finding logic to multivariate vector valued functions by modifying the intensity function using Equation 16, which involves a norm like f(x).
-
AI systems can identify all points of extrema for a function by finding the zeroes of its derivative, or in the multivariate case, by applying PPZ to approximate the gradient using Equation 20 to find solutions of Gradient f(x) = 0.
-
The Adaptive PPPZ (APPPZ) algorithm allows AI systems to achieve a desired end tolerance T end by refining the search process iteratively, which can take a "lot of time to run for T end < 1e-20, depending on the function under consideration and the chosen values for parameters K and Q."
-
The Cox point process approach allows AI systems to replace the independence assumption of Poisson point processes with conditional independence by treating K as a random variable, leading to a
random intensity (a random field).
Related papers
- Learning Robust Treatment Rules for Censored Data
- Neural Estimation of Summary Statistics for Simulation-Based Inference with Non-Asymptotic Guarantees
- Spectral divide-and-conquer MCMC for long stationary time series
- Improving the Efficiency of Subgroup Analysis in Randomized Controlled Trials with TMLE
- Personalized Federated Vector Autoregression with Personalization Diversity
- Doubly Robust and Efficient Calibration of Prediction Sets for Right-Censored Time-to-Event Outcomes