March sessions + discarded assets

Author: k-sidorov

_bibliography/s20250217.bib

@@ -0,0 +1,26 @@
+@InProceedings{10.1007/978-3-031-26409-2_13,
+author="Vos, Dani{\"e}l
+and Verwer, Sicco",
+editor="Amini, Massih-Reza
+and Canu, St{\'e}phane
+and Fischer, Asja
+and Guns, Tias
+and Kralj Novak, Petra
+and Tsoumakas, Grigorios",
+title="Adversarially Robust Decision Tree Relabeling",
+booktitle="Machine Learning and Knowledge Discovery in Databases",
+year="2023",
+publisher="Springer Nature Switzerland",
+address="Cham",
+pages="203--218",
+abstract="Decision trees are popular models for their interpretation properties and their success in ensemble models for structured data. However, common decision tree learning algorithms produce models that suffer from adversarial examples. Recent work on robust decision tree learning mitigates this issue by taking adversarial perturbations into account during training. While these methods generate robust shallow trees, their relative quality reduces when training deeper trees due the methods being greedy. In this work we propose robust relabeling, a post-learning procedure that optimally changes the prediction labels of decision tree leaves to maximize adversarial robustness. We show this can be achieved in polynomial time in terms of the number of samples and leaves. Our results on 10 datasets show a significant improvement in adversarial accuracy both for single decision trees and tree ensembles. Decision trees and random forests trained with a state-of-the-art robust learning algorithm also benefited from robust relabeling.",
+isbn="978-3-031-26409-2",
+  html="https://link.springer.com/chapter/10.1007/978-3-031-26409-2_13"
+}
+
+
+@article{Vos_Verwer_2022,
+  title={Robust Optimal Classification Trees against Adversarial Examples}, volume={36},
+  url={https://ojs.aaai.org/index.php/AAAI/article/view/20829},
+  html={https://ojs.aaai.org/index.php/AAAI/article/view/20829},
+  DOI={10.1609/aaai.v36i8.20829}, abstractNote={Decision trees are a popular choice of explainable model, but just like neural networks, they suffer from adversarial examples. Existing algorithms for fitting decision trees robust against adversarial examples are greedy heuristics and lack approximation guarantees. In this paper we propose ROCT, a collection of methods to train decision trees that are optimally robust against user-specified attack models. We show that the min-max optimization problem that arises in adversarial learning can be solved using a single minimization formulation for decision trees with 0-1 loss. We propose such formulations in Mixed-Integer Linear Programming and Maximum Satisfiability, which widely available solvers can optimize. We also present a method that determines the upper bound on adversarial accuracy for any model using bipartite matching. Our experimental results demonstrate that the existing heuristics achieve close to optimal scores while ROCT achieves state-of-the-art scores.}, number={8}, journal={Proceedings of the AAAI Conference on Artificial Intelligence}, author={Vos, Daniël and Verwer, Sicco}, year={2022}, month={Jun.}, pages={8520-8528} }

_bibliography/s20250224.bib

@@ -0,0 +1,15 @@
+@article{KARADEMIR2025480,
+title = {A two-echelon multi-trip vehicle routing problem with synchronization for an integrated water- and land-based transportation system},
+journal = {European Journal of Operational Research},
+volume = {322},
+number = {2},
+pages = {480-499},
+year = {2025},
+issn = {0377-2217},
+doi = {https://doi.org/10.1016/j.ejor.2024.10.047},
+url = {https://www.sciencedirect.com/science/article/pii/S0377221724008506},
+html = {https://www.sciencedirect.com/science/article/pii/S0377221724008506},
+author = {Cigdem Karademir and Breno A. Beirigo and Bilge Atasoy},
+keywords = {Two-echelon network, Vehicle routing problems, City logistics, Satellite synchronization, Logic-based Benders’ decomposition},
+abstract = {This study focuses on two-echelon synchronized logistics problems in the context of integrated water- and land-based transportation (IWLT) systems. The aim is to meet the increasing demand in city logistics as a result of the growth in transport activities, including parcel delivery, food delivery, and waste collection. We propose two models, a novel mixed integer linear joint model, and a logic-based Benders’ decomposition (LBBD) model, for a two-echelon problem under realistic settings such as multi-trips, time windows, and synchronization at the satellites with no storage and limited resource capacities. The objective is to optimize transfers and satellite assignments, thereby reducing overall logistics costs for street vehicles and vessels. Computational experiments demonstrate that the LBBD model is more robust in terms of solution quality and solution time on average while the added value of the LBBD is more evident when solving large-scale instances with 100 customers, reducing the overall costs by 10.6% on average and significantly reducing the fleet costs on both networks. Furthermore, we assess the effect of changing cost parameters and satellite locations in the proposed IWLT system–analyzing system behavior and suggesting potential improvements–and evaluate several system alternatives in city logistics–consisting of different transportation network designs (single- and two-echelon), vehicle types, and operational constraints. On average, the proposed two-echelon IWLT system reduces the number of kilometers traveled by vehicles at street level by ranging from 20% to 30% compared to a typical single-echelon service design that relies solely on trucks.}
+}

_bibliography/s20250303.bib

@@ -0,0 +1,25 @@
+@article{HYAFIL197615,
+title = {Constructing optimal binary decision trees is NP-complete},
+journal = {Information Processing Letters},
+volume = {5},
+number = {1},
+pages = {15-17},
+year = {1976},
+issn = {0020-0190},
+doi = {https://doi.org/10.1016/0020-0190(76)90095-8},
+url = {https://www.sciencedirect.com/science/article/pii/0020019076900958},
+html = {https://www.sciencedirect.com/science/article/pii/0020019076900958},
+author = {Laurent Hyafil and Ronald L. Rivest},
+keywords = {Binary decision trees, computational complexity, NP-complete}
+}
+
+  @misc{staus2024wittyefficientsolvercomputing,
+      title={Witty: An Efficient Solver for Computing Minimum-Size Decision Trees}, 
+      author={Luca Pascal Staus and Christian Komusiewicz and Frank Sommer and Manuel Sorge},
+      year={2024},
+      eprint={2412.11954},
+      archivePrefix={arXiv},
+      primaryClass={cs.DS},
+      url={https://arxiv.org/abs/2412.11954}, 
+      html={https://arxiv.org/abs/2412.11954}, 
+}

_bibliography/s20250310.bib

@@ -0,0 +1,17 @@
+@article{Pilanci2015,
+  title = {Sparse learning via Boolean relaxations},
+  volume = {151},
+  ISSN = {1436-4646},
+  url = {http://dx.doi.org/10.1007/s10107-015-0894-1},
+  html = {http://dx.doi.org/10.1007/s10107-015-0894-1},
+  DOI = {10.1007/s10107-015-0894-1},
+  number = {1},
+  journal = {Mathematical Programming},
+  publisher = {Springer Science and Business Media LLC},
+  author = {Pilanci,  Mert and Wainwright,  Martin J. and El Ghaoui,  Laurent},
+  year = {2015},
+  month = mar,
+  pages = {63–87}
+}
+
+

_bibliography/s20250317.bib

@@ -0,0 +1,14 @@
+@InProceedings{10.1007/978-3-642-04244-7_62,
+author="Vil{\'i}m, Petr",
+editor="Gent, Ian P.",
+title="Edge Finding Filtering Algorithm for Discrete Cumulative Resources in $O(kn \log n)$",
+booktitle="Principles and Practice of Constraint Programming - CP 2009",
+year="2009",
+publisher="Springer Berlin Heidelberg",
+address="Berlin, Heidelberg",
+pages="802--816",
+abstract="This paper presents new Edge Finding algorithm for discrete cumulative resources, i.e. resources which can process several activities simultaneously up to some maximal capacity C. The algorithm has better time complexity than the current version of this algorithm: {\$}{\{}{\backslash}mathcal O{\}}(kn {\{}{\backslash}rm log{\}} n){\$}versus {\$}{\{}{\backslash}mathcal O{\}}(k n^2){\$}where n is number of activities on the resource and k is number of distinct capacity demands. Moreover the new algorithm is slightly stronger and it is able to handle optional activities. The algorithm is based on the $\Theta$-tree -- a binary tree data structure which already proved to be very useful in filtering algorithms for unary resource constraints.",
+isbn="978-3-642-04244-7",
+html="https://link.springer.com/chapter/10.1007/978-3-642-04244-7_62"
+}
+

_bibliography/s20250407.bib

@@ -0,0 +1,15 @@
+@article{Achterberg2009,
+  title = {SCIP: solving constraint integer programs},
+  volume = {1},
+  ISSN = {1867-2957},
+  url = {http://dx.doi.org/10.1007/s12532-008-0001-1},
+  html = {http://dx.doi.org/10.1007/s12532-008-0001-1},
+  DOI = {10.1007/s12532-008-0001-1},
+  number = {1},
+  journal = {Mathematical Programming Computation},
+  publisher = {Springer Science and Business Media LLC},
+  author = {Achterberg,  Tobias},
+  year = {2009},
+  month = jan,
+  pages = {1–41}
+}

_config.yml

@@ -161,13 +161,13 @@ disqus_shortname: al-folio # put your disqus shortname
 # External sources.
 # If you have blog posts published on medium.com or other external sources,
 # you can display them in your blog by adding a link to the RSS feed.
-external_sources:
-  - name: medium.com
-    rss_url: https://medium.com/@al-folio/feed
-  - name: Google Blog
-    posts:
-      - url: https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/
-        published_date: 2024-05-14
+# external_sources:
+#   - name: medium.com
+#     rss_url: https://medium.com/@al-folio/feed
+#   - name: Google Blog
+#     posts:
+#       - url: https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/
+#         published_date: 2024-05-14
 
 # -----------------------------------------------------------------------------
 # Newsletter

_sessions/2025-02-17-robust-decision-trees.md

@@ -2,11 +2,12 @@
 layout: session
 title: Robust Decision Trees
 host: Daniël Vos
+host_ref: https://daniel-vos.github.io/
 session_type: research
+related_papers: s20250217
 ---
 
 Decision trees are a popular choice of interpretable model, but just like neural networks, they suffer from adversarial examples: adversarial perturbations that result in misclassifications. Existing algorithms for fitting decision
- trees that are robust against adversarial examples are greedy heuristics and, therefore, lack approximation guarantees. We propose Robust Optimal Classification Trees (ROCT), a collection of methods to train decision trees that are optimally robust against
- user-specified attack models. We show that the min-max optimization problem that arises in adversarial learning can be solved using a single minimization formulation for decision trees with 0-1 loss. We propose such formulations in Mixed-Integer Linear Programming
- and Maximum Satisfiability, which widely available solvers can optimize. We also present a method that uses bipartite matching to compute robust leaf labels in polynomial time and can be used to determine the upper bound on adversarial accuracy for any model.
- 
+trees that are robust against adversarial examples are greedy heuristics and, therefore, lack approximation guarantees. We propose Robust Optimal Classification Trees (ROCT), a collection of methods to train decision trees that are optimally robust against
+user-specified attack models. We show that the min-max optimization problem that arises in adversarial learning can be solved using a single minimization formulation for decision trees with 0-1 loss. We propose such formulations in Mixed-Integer Linear Programming
+and Maximum Satisfiability, which widely available solvers can optimize. We also present a method that uses bipartite matching to compute robust leaf labels in polynomial time and can be used to determine the upper bound on adversarial accuracy for any model.

_sessions/2025-02-24-benders.md

@@ -0,0 +1,12 @@
+---
+layout: session
+title: Robust Decision Trees
+host: Çiğdem Karademir
+session_type: research
+related_papers: s20250224
+---
+
+In this talk, we address a real-world problem in city logistics: the Two-Echelon Multi-Trip Vehicle Routing Problem with Satellite Synchronization (2E-MVRP-SS). The problem involves coordinating a fleet of light electric freight vehicles (LEFVs) operating at the street level with a fleet of vessels operating at the water level. The objective is to minimize total logistics costs while serving customer requests with time windows and transshipping goods between LEFVs and vessels at satellite locations. These satellites have limited capacity and no storage, necessitating precise synchronization of operations in both space and time.
+
+To tackle this problem, we propose two models: a joint mixed-integer linear programming (MILP) model and a logic-based Benders decomposition (LBBD) model. The LBBD model demonstrates greater robustness for larger instances, significantly improving solution quality and computational efficiency, with an 18.7% reduction in total travel time. We also investigate the impact of cost allocations between two service providers and satellite locations on overall system performance. Furthermore, we evaluate various service design alternatives, such as single echelon systems using only trucks, only LEFVs, and two-echelon systems using only storage, and synchronization.
+The key contribution of this work lies in the development of efficient models for solving the 2E-MVRP-SS, offering valuable insights into the trade-offs between different system configurations and operational parameters. This research is highly relevant to the fields of scheduling and optimization, providing practical solutions for advancing sustainable urban logistics.

_sessions/2025-03-03-perfect-trees.md

@@ -0,0 +1,9 @@
+---
+layout: session
+title: Perfect Decision Trees of Minimum Size
+host: Koos van der Linden
+session_type: paper
+related_papers: s20250303
+---
+
+We address the challenge of developing efficient Constraint Programming-based approaches for solving formulas over the quantifier-free fragment of the theory of bitvectors (BV), which is of paramount importance in software verification. We propose CP(BV), a highly efficient BV resolution technique built on carefully chosen anterior results sharpened with key original features such as thorough domain combination or dedicated labeling. Extensive experimental evaluations demonstrate that CP(BV) is much more efficient than previous similar attempts from the CP community, that it is indeed able to solve the majority of the standard verification benchmarks for bitvectors, and that it already complements the standard SMT approaches on several crucial (and industry-relevant) aspects, notably in terms of scalability w.r.t. bit-width, theory combination or intricate mix of non-linear arithmetic and bitwise operators. This work paves the way toward building competitive CP-based verification-oriented solvers.