Juha Harviainen
@juhahar.bsky.social
Postdoctoral researcher at University of Helsinki / parameterized algorithms, perfect sampling, and Bayesian networks / https://juhaharviainen.com/
Reposted by Juha Harviainen
This is by a large margin the most serious problem and mistake in conference peer review I have seen in my career. Apparently a many people were aware of this and many could find out who their reviewers were. This probably has created a large number of unnecessary enmities.
@iclr-conf.bsky.social
@iclr-conf.bsky.social
November 27, 2025 at 7:45 PM
This is by a large margin the most serious problem and mistake in conference peer review I have seen in my career. Apparently a many people were aware of this and many could find out who their reviewers were. This probably has created a large number of unnecessary enmities.
@iclr-conf.bsky.social
@iclr-conf.bsky.social
Reposted by Juha Harviainen
HIIT Open 2025 programming contest has started!
November 8, 2025 at 10:17 AM
HIIT Open 2025 programming contest has started!
Reposted by Juha Harviainen
HIIT Open 2025 programming contest coming on Saturday, 45 contestants in 20 teams competing at Aalto University!
open.hiit.fi
open.hiit.fi
HIIT Open 2025 Programming Contest
HIIT Open 2025 Programming Contest, Saturday, 8 Nov 2025, Otaniemi, Espoo, Finland.
open.hiit.fi
November 6, 2025 at 5:18 PM
HIIT Open 2025 programming contest coming on Saturday, 45 contestants in 20 teams competing at Aalto University!
open.hiit.fi
open.hiit.fi
New preprint on arXiv: arxiv.org/abs/2510.25254
We propose two novel improvements to MCMC for Bayesian network structures by speeding up the basic moves and pruning less relevant parent sets, both resulting in one to three orders of magnitude improvements in the running time.
We propose two novel improvements to MCMC for Bayesian network structures by speeding up the basic moves and pruning less relevant parent sets, both resulting in one to three orders of magnitude improvements in the running time.
Scaling Up Bayesian DAG Sampling
Bayesian inference of Bayesian network structures is often performed by sampling directed acyclic graphs along an appropriately constructed Markov chain. We present two techniques to improve sampling....
arxiv.org
October 30, 2025 at 9:17 AM
New preprint on arXiv: arxiv.org/abs/2510.25254
We propose two novel improvements to MCMC for Bayesian network structures by speeding up the basic moves and pruning less relevant parent sets, both resulting in one to three orders of magnitude improvements in the running time.
We propose two novel improvements to MCMC for Bayesian network structures by speeding up the basic moves and pruning less relevant parent sets, both resulting in one to three orders of magnitude improvements in the running time.
Reposted by Juha Harviainen
HIIT Open 2025 coming on November 8, the registration form is now open! open.hiit.fi
HIIT Open 2025 Programming Contest
HIIT Open 2025 Programming Contest, Saturday, 8 Nov 2025, Otaniemi, Espoo, Finland.
open.hiit.fi
October 10, 2025 at 9:02 AM
HIIT Open 2025 coming on November 8, the registration form is now open! open.hiit.fi
Reposted by Juha Harviainen
New papers from the HALT community: 2 x SODA 2025, 2 x NeurIPS 2025, 3 x Asiacrypt 2025, and more!
algorithms.fi/publications/
algorithms.fi/publications/
HALT: Publications
algorithms.fi
October 6, 2025 at 1:26 PM
New papers from the HALT community: 2 x SODA 2025, 2 x NeurIPS 2025, 3 x Asiacrypt 2025, and more!
algorithms.fi/publications/
algorithms.fi/publications/
Our follow-up paper to this, "Improving Decision Trees through the Lens of Parameterized Local Search", with Sommer and Sorge, got accepted to NeurIPS '25! We study the parameterized complexity of improving decision trees by modifying multiple internal nodes simultaneously to avoid local optimas.
With colleagues from TU Wien—Frank Sommer, Manuel Sorge, and @szeider.bsky.social—our paper "Optimal Decision Tree Pruning Revisited: Algorithms and Complexity" got accepted to ICML. The paper provides a comprehensive study of parameterized complexity of pruning decision trees to reduce overfitting.
September 22, 2025 at 6:33 PM
Our follow-up paper to this, "Improving Decision Trees through the Lens of Parameterized Local Search", with Sommer and Sorge, got accepted to NeurIPS '25! We study the parameterized complexity of improving decision trees by modifying multiple internal nodes simultaneously to avoid local optimas.
Reposted by Juha Harviainen
Most of the presentation slides from Theory Days now available online: algorithms.fi/theory-days-...
Helsinki Algorithms & Theory Days
algorithms.fi
September 5, 2025 at 10:11 AM
Most of the presentation slides from Theory Days now available online: algorithms.fi/theory-days-...
Reposted by Juha Harviainen
Helsinki Algorithms & Theory Days program is now available online: algorithms.fi/theory-days-...
Helsinki Algorithms & Theory Days
algorithms.fi
August 17, 2025 at 8:36 AM
Helsinki Algorithms & Theory Days program is now available online: algorithms.fi/theory-days-...
Reposted by Juha Harviainen
Helsinki Algorithms & Theory Days on 28–29 August, 2025, keynote talk by Andris Ambainis
algorithms.fi/theory-days-...
algorithms.fi/theory-days-...
Helsinki Algorithms & Theory Days
algorithms.fi
July 24, 2025 at 8:51 PM
Helsinki Algorithms & Theory Days on 28–29 August, 2025, keynote talk by Andris Ambainis
algorithms.fi/theory-days-...
algorithms.fi/theory-days-...
Preprint now available on arXiv:
arxiv.org/abs/2305.19673
arxiv.org/abs/2305.19673
In collaboration with Kseniya Rychkova and Mikko Koivisto, our paper "Quantum Speedups for Bayesian Network Structure Learning" was accepted to UAI 2025.
We show the time complexity of classical algorithms for structure learning to be presumably optimal and give better quantum algorithms.
We show the time complexity of classical algorithms for structure learning to be presumably optimal and give better quantum algorithms.
June 3, 2025 at 7:23 AM
Preprint now available on arXiv:
arxiv.org/abs/2305.19673
arxiv.org/abs/2305.19673
In collaboration with Kseniya Rychkova and Mikko Koivisto, our paper "Quantum Speedups for Bayesian Network Structure Learning" was accepted to UAI 2025.
We show the time complexity of classical algorithms for structure learning to be presumably optimal and give better quantum algorithms.
We show the time complexity of classical algorithms for structure learning to be presumably optimal and give better quantum algorithms.
May 7, 2025 at 1:48 PM
In collaboration with Kseniya Rychkova and Mikko Koivisto, our paper "Quantum Speedups for Bayesian Network Structure Learning" was accepted to UAI 2025.
We show the time complexity of classical algorithms for structure learning to be presumably optimal and give better quantum algorithms.
We show the time complexity of classical algorithms for structure learning to be presumably optimal and give better quantum algorithms.
With colleagues from TU Wien—Frank Sommer, Manuel Sorge, and @szeider.bsky.social—our paper "Optimal Decision Tree Pruning Revisited: Algorithms and Complexity" got accepted to ICML. The paper provides a comprehensive study of parameterized complexity of pruning decision trees to reduce overfitting.
May 2, 2025 at 6:01 AM
With colleagues from TU Wien—Frank Sommer, Manuel Sorge, and @szeider.bsky.social—our paper "Optimal Decision Tree Pruning Revisited: Algorithms and Complexity" got accepted to ICML. The paper provides a comprehensive study of parameterized complexity of pruning decision trees to reduce overfitting.
The paper is now available in AISTATS'25 proceedings:
proceedings.mlr.press/v258/harviai...
proceedings.mlr.press/v258/harviai...
Our paper "On Tractability of Learning Bayesian Networks with Ancestral Constraints" with Pekka Parviainen was accepted to AISTATS'25 🎉
The paper studies parameterized complexity of structure learning for varying constraints when we can say something about the order of any set of nodes larger than k
The paper studies parameterized complexity of structure learning for varying constraints when we can say something about the order of any set of nodes larger than k
April 25, 2025 at 1:29 PM
The paper is now available in AISTATS'25 proceedings:
proceedings.mlr.press/v258/harviai...
proceedings.mlr.press/v258/harviai...
Reposted by Juha Harviainen
Today's Frontier Friday material here: algorithms.fi/frontier-fri...
March 14, 2025 at 1:01 PM
Today's Frontier Friday material here: algorithms.fi/frontier-fri...
Reposted by Juha Harviainen
Information on HIIT Frontier Friday now here, next meeting: 14 Mar 2025
algorithms.fi/frontier-fri...
algorithms.fi/frontier-fri...
HIIT Frontier Friday
algorithms.fi
February 19, 2025 at 6:49 PM
Information on HIIT Frontier Friday now here, next meeting: 14 Mar 2025
algorithms.fi/frontier-fri...
algorithms.fi/frontier-fri...
Our paper "On Tractability of Learning Bayesian Networks with Ancestral Constraints" with Pekka Parviainen was accepted to AISTATS'25 🎉
The paper studies parameterized complexity of structure learning for varying constraints when we can say something about the order of any set of nodes larger than k
The paper studies parameterized complexity of structure learning for varying constraints when we can say something about the order of any set of nodes larger than k
January 23, 2025 at 10:57 AM
Our paper "On Tractability of Learning Bayesian Networks with Ancestral Constraints" with Pekka Parviainen was accepted to AISTATS'25 🎉
The paper studies parameterized complexity of structure learning for varying constraints when we can say something about the order of any set of nodes larger than k
The paper studies parameterized complexity of structure learning for varying constraints when we can say something about the order of any set of nodes larger than k