[go: up one dir, main page]

Archive for ODE

more than mostly MC

Posted in Kids, pictures, Statistics, Travel, University life with tags , , , , , , , , , , , , , , , , , , , , , , on December 27, 2024 by xi'an

The session of last Friday (and last one of 2024!) proved most interesting, with two (fully) Monte Carlo talks. The first one by Louis Grenioux was about an improvement on diffusion sampler, following a recent arXival by Maxime Noble and co-authors. Which brought me back to the long-standing multimodal challenge in Monte Carlo methods. When all modes of a target distribution are known, even roughly, this is not much of an issue since samplers can be arm-bent into visiting all these modes. But the problem becomes much harder when the location and a fortiori the number of modes are not known. The paper aims at adapting diffusion samplers towards a better exploration of the modes, albeit their location is known. Otherwise, using MCMC as a starting (reference) distribution would risk missing some of them. Which also explains why the authors can rely on the classical Gaussian mixtures proposal as a cheap substitute to neural networks (EBM). Since, within diffusion models, both intermediary distributions and their scores are intractable, they also introduce a variational parametric approximation that can be optimised.

The second talk was given by Guillaume Chennetier, in connection with his recent PhD thesis, developing a form of X-entropy sampling for rare events. As in nuclear plant major accidents. It took me a while to realise that PDMPs were not used as a simulation tool, as in the zigzag sampler and its avatars, The proposal involved creating a graph structure on the space of PDMP trajectories and designing the optimal importance process (yes, the one with zero variance!) using so-called committor functions that modify jump intensity and kernel, in a sequential way reminiscent of X-entropy. The approach recycles past trajectories as Monte Carlo elements if missing an adaptive mixture importance sampling (AMIS!) version that would bring more stability. The talk also included an interesting pointer to the availability of the distribution of the PDMP path, thus treated as a likelihood. (!). The signage at the entrance of the Monte-Carlo (mind the hyphen!) casino also made an appearance, reminding me of our memorable group picture on the same spot, eons ago! (But not of whom took the picture!)

Leave a comment »

6th Workshop on Sequential Monte Carlo Methods

Posted in Mountains, pictures, Statistics, Travel, University life with tags , , , , , , , , , , , , , , , , , , , , , , , , , , on May 16, 2024 by xi'an

Very glad to be back to an SMC workshop as it has been nine years since my attending SMC 2015 in Malakoff! The more for the workshop taking place in Edinburgh and at the Bayes Centre. It is one of these places where I feel somewhat returning to familiar grounds with accumulated memories. Like my last visit there when I had a tea with Mike Titterington…

The overall pace of the workshop was quite nice, with long breaks for informal discussions (and time for ‘oggin’!) and interesting poster late afternoons, helped by the small number of them at each instance, incl. one on reversible jump HMC. Here are a few scribbled entries about some talks along the first two days.

After my opening talk (!), Joaquín Míguez talked about the impact of a sequential (Euler-Marayama) discretisation scheme for stochastic differential equations on Bayesian filtering with control of the approximation effect. Axel Finke (in a joint work with Adrien Corenflos, now an ERC Ocean postdoc in Warwick) built a sequence of particle filter algorithms targeting good performances (high expected jumping distance) against both large dimensions and high time horizon, exploiting gradient shift MALA-like, as well as prior impact, with the conclusion that their jack-of-all-trades solutions, Particle­-MALA and Particle­-mGRAD, enjoyed this resistance in nearly normal models. Interesting reminder of the auxiliary particle trick and good insights on using the smoothing target, even when accounting for the computing time, but too many versions for a single talk without checking against the preprint.

The SMC sampler-like algorithm involves propagating N “seed” particles z(i), with a mutation mechanism consisting of the generation of N integrator snippets 𝗓:=(z,ψ⁢(z),ψ²⁢(z),…) started at every seed particle z(i), resulting in N×(T+1) particles which are then whittled down to a set of N seed particles using a standard resampling scheme. Andrieu et al., 2024

Christophe Andrieu talked about Monte Carlo sampling with integrator snippets, starting with recycling solutions for the leapfrog integrator HMC and unfolding Hamiltonians for moving more easily. With snippets representing discretised paths along the level sets being used as particles, picking zero, one, or more particles along each path, since importance weights are connection with multinomial HMC

This relatively small algorithmic modification of the conditional particle filter, which we call the conditional backward sampling particle filter has a dramatically improved performance over the conditional particle filter. Karjalainen et al., 2024

Anthony Lee looked at mixing times for backward sampling SMC (CBPF/ancestor sampling) cf Lee et al. (2020), where the backward step consists in computing the weight of a randomly drawn backward or ancestral history. Improving on earlier results to reach mixing time O(log T) and complexity O(T log T) (with T the time horizon). Thanks to maximal coupling and boundedness assumptions on the prior and likelihood functions.

Neil Chada presented a work on Bayesian multilevel Monte Carlo on deep networks. À la Giles, with a telescoping identity. Always puzzling to envision a prior on all parameters of a neural network. Achieving a computational cost inverse to the order of the MSE, at best. With a useful reminder that pushing the size of the NN to infinity results in a (poor) Gaussian process prior (Sell et al., 2023).

On my first evening, I stopped with a friend in my favourite Blonde [restaurant], as in almost every other visit to Edinburgh, enjoyable as always, but I also found the huge offer of Asian minimarkets in the area too tempting to resist, between Indian, Korean, and Chinese products. (Although with a disappointing hojicha!). As I could not reach any new Munro by train or bus within a reasonable time range I resorted to the nearer Pentland Hills, with a stop by Rosslyn Chapel (mostly of Da Vinci Code fame!, if classic enough). And some delays in finding a bus getting there (misled by google map!) and a trail (misled by my poor map reading skills) up the actual hills. The mist did not help either.

Leave a comment »

invertible flow non equilibrium sampling (InFiNE)

Posted in Books, Statistics, University life with tags , , , , , , , , , , , , , on May 21, 2021 by xi'an

With Achille Thin and a few other coauthors [and friends], we just arXived a paper on a new form of importance sampling, motivated by a recent paper of Rotskoff and Vanden-Eijnden (2019) on non-equilibrium importance sampling. The central ideas of this earlier paper are the introduction of conformal Hamiltonian dynamics, where a dissipative term is added to the ODE found in HMC, namely

\dfrac{\text d p_t}{\text dt}=-\dfrac{\partial}{\partial q}H(q_t,p_t)-\gamma p_t=-\nabla U(q_t)-\gamma p_t

which means that all orbits converge to fixed points that satisfy ∇U(q) = 0 as the energy eventually vanishes. And the property that, were T be a conformal Hamiltonian integrator associated with H, i.e. perserving the invariant measure, averaging over orbits of T would improve the precision of Monte Carlo unbiased estimators, while remaining unbiased. The fact that Rotskoff and Vanden-Eijnden (2019) considered only continuous time makes their proposal hard to implement without adding approximation error, while our approach is directly set in discrete-time and preserves unbiasedness. And since measure preserving transforms are too difficult to come by, a change of variable correction, as in normalising flows, allows for an arbitrary choice of T, while keeping the estimator unbiased. The use of conformal maps makes for a natural choice of T in this context.

The resulting InFiNE algorithm is an MCMC particular algorithm which can be represented as a  partially collapsed Gibbs sampler when using the right auxiliary variables. As in Andrieu, Doucet and Hollenstein (2010) and their ISIR algorithm. The algorithm can be used for estimating normalising constants, comparing favourably with AIS, sampling from complex targets, and optimising variational autoencoders and their ELBO.

I really appreciated working on this project, with links to earlier notions like multiple importance sampling à la Owen and Zhou (2000), nested sampling, non-homogeneous normalising flows, measure estimation à la Kong et al. (2002), on which I worked in a more or less distant past.

Leave a comment »

IMS workshop [day 3]

Posted in pictures, R, Statistics, Travel, University life with tags , , , , , , , , , , , , , , , , , , , on August 30, 2018 by xi'an

I made the “capital” mistake of walking across the entire NUS campus this morning, which is quite green and pretty, but which almost enjoys an additional dimension brought by such an intense humidity that one feels having to get around this humidity!, a feature I have managed to completely erase from my memory of my previous visit there. Anyway, nothing of any relevance. oNE talk in the morning was by Markus Eisenbach on tools used by physicists to speed up Monte Carlo methods, like the Wang-Landau flat histogram, towards computing the partition function, or the distribution of the energy levels, definitely addressing issues close to my interest, but somewhat beyond my reach for using a different language and stress, as often in physics. (I mean, as often in physics talks I attend.) An idea that came out clear to me was to bypass a (flat) histogram target and aim directly at a constant slope cdf for the energy levels. (But got scared away by the Fourier transforms!)

Lawrence Murray then discussed some features of the Birch probabilistic programming language he is currently developing, especially a fairly fascinating concept of delayed sampling, which connects with locally-optimal proposals and Rao Blackwellisation. Which I plan to get back to later [and hopefully sooner than later!].

In the afternoon, Maria de Iorio gave a talk about the construction of nonparametric priors that create dependence between a sequence of functions, a notion I had not thought of before, with an array of possibilities when using the stick breaking construction of Dirichlet processes.

And Christophe Andrieu gave a very smooth and helpful entry to partly deterministic Markov processes (PDMP) in preparation for talks he is giving next week for the continuation of the workshop at IMS. Starting with the guided random walk of Gustafson (1998), which extended a bit later into the non-reversible paper of Diaconis, Holmes, and Neal (2000). Although I had a vague idea of the contents of these papers, the role of the velocity ν became much clearer. And premonitory of the advances made by the more recent PDMP proposals. There is obviously a continuation with the equally pedagogical talk Christophe gave at MCqMC in Rennes two months [and half the globe] ago,  but the focus being somewhat different, it really felt like a new talk [my short term memory may also play some role in this feeling!, as I now remember the discussion of Hilderbrand (2002) for non-reversible processes]. An introduction to the topic I would recommend to anyone interested in this new branch of Monte Carlo simulation! To be followed by the most recently arXived hypocoercivity paper by Christophe and co-authors.

1 Comment »

Sequentially Constrained Monte Carlo

Posted in Books, Mountains, pictures, Statistics, University life with tags , , , , , , , , , , on November 7, 2014 by xi'an

This newly arXived paper by S. Golchi and D. Campbell from Vancouver (hence the above picture) considers the (quite) interesting problem of simulating from a target distribution defined by a constraint. This is a question that have bothered me for a long while as I could not come up with a satisfactory solution all those years… Namely, when considering a hard constraint on a density, how can we find a sequence of targets that end up with the restricted density? This is of course connected with the zero measure case posted a few months ago. For instance, how do we efficiently simulate a sample from a Student’s t distribution with a fixed sample mean and a fixed sample variance?

“The key component of SMC is the filtering sequence of distributions through which the particles evolve towards the target distribution.” (p.3)

This is indeed the main issue! The paper considers using a sequence of intermediate targets hardening progressively the constraint(s), along with an SMC sampler, but this recommendation remains rather vague and hence I am at loss as to how to make it work when the exact constraint implies a change of measure. The first example is monotone regression where y has mean f(x) and f is monotone. (Everything is unidimensional here.) The sequence is then defined by adding a multiplicative term that is a function of ∂f/∂x, for instance

Φ(τ∂f/∂x),

with τ growing to infinity to make the constraint moving from soft to hard. An interesting introduction, even though the hard constraint does not imply a change of parameter space or of measure. The second example is about estimating the parameters of an ODE, with the constraint being the ODE being satisfied exactly. Again, not exactly what I was looking for. But with an exotic application to deaths from the 1666 Black (Death) plague.

And then the third example is about ABC and the choice of summary statistics! The sequence of constraints is designed to keep observed and simulated summary statistics close enough when the dimension of those summaries increases, which means they are considered simultaneously rather than jointly. (In the sense of Ratmann et al., 2009. That is, with a multidimensional distance.) The model used for the application of the SMC is the dynamic model of Wood (2010, Nature). The outcome of this specific implementation is not that clear compared with alternatives… And again sadly does not deal with the/my zero measure issue.

Leave a comment »