Sign in

Alexander Helber

@ahelber.bsky.social
41 followers 70 following 63 posts

PhD student at the Chair of Operations Research @ RWTH Aachen

PostsRepliesMedia
Reposted by Alexander Helber
Wehrlegig Games @wehrlegig.games · 06/10/2026
A cutthroat business game about the dramatic rise of the opium trade in 19th-century China. Live now on Kickstarter. www.kickstarter.com/projects/col...
kickstarter.com
An Infamous Traffic: Second Edition
A new edition of the acclaimed game of cutthroat capitalism in the nineteenth century opium trade.
37613
Alexander Helber @ahelber.bsky.social · 06/10/2026
Cool cool, so a company can now decide who to "give" a result to? Feels weird. (Without any judgment on anyone involved in this particular case.)
000
Reposted by Alexander Helber
Matteo Fischetti @matteofischetti.bsky.social · 30/09/2026
Just a visual presentation of EndoWake (joint work with Martina Fischetti)
021
Reposted by Alexander Helber
Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 30/09/2026
In our Nature paper, we introduce the first superhuman Stratego AI, which we built using general techniques that we developed for RL & test-time compute under imperfect information. www.nature.com/articles/s41...
nature.com
Scalable decision-making for games of imperfect information - Nature
Ataraxos, an AI for the board wargame Stratego, establishes a design pattern for reinforcement learning and search that is effective under large amounts of hidden information, a longstanding desiderat...
1225960
Alexander Helber @ahelber.bsky.social · 29/09/2026
Today: + Ready to submit new paper - Paper from last year needs to undergo a *third* major revision Would have been nice to celebrate today and get the news tomorrow...
000
Reposted by Alexander Helber
Clément Canonne @ccanonne.github.io · 23/09/2026
Looking for takes by TCS (Theoretical Computer Science) senior academics/researchers about AI and their discipline/research values. E.g., essays, detailed blog posts. (Specifically TCS: there is a lot from Maths, but while adjacent there are enough differences in culture it's worth distinguishing)
63110
Reposted by Alexander Helber
Clément Canonne @ccanonne.github.io · 19/09/2026
Third paper on ≈the same open question within days: this one is the third. The authors blatantly acknowledge their whole plan was to get and dump a ChatGPT proof on arXiv in ONE day to claim a result, and complain about being "scooped" by another group WHICH DID THE SAME THING Do you feel no shame?
5457
Alexander Helber @ahelber.bsky.social · 06/09/2026
Which tricks would these be, apart from anki? (agreed btw SR is the only sane way if large amounts of material need to be committed to memory)
000
Alexander Helber @ahelber.bsky.social · 28/08/2026
Which makes one wonder, if you want to propose a new algorithm in the future, what is the baseline? Do you have to query every top-of-the-line model (how often?) and compare against their algorithms to demonstrate that your proposal is good?
000
Reposted by Alexander Helber
Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 26/08/2026
I am begging you, stop using LLMs to write your papers it is horrible to read and causes an instant DNF.
71349
Alexander Helber @ahelber.bsky.social · 27/07/2026
This post made me look up the books, now finished with the fourth - thanks!
110
Alexander Helber @ahelber.bsky.social · 11/07/2026
I think for small-ish courses there is still potential for letting students do work at home and then explain or discuss it in person. But yeah generally very annoying that everything you can't directly observe has now 0 signal to noise
000
Alexander Helber @ahelber.bsky.social · 07/07/2026
The Claude-speak turned to 11 does hurt my braincells though
010
Reposted by Alexander Helber
Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 01/07/2026
There really should be a lot more shame attached to working for an Elon company
4896
Alexander Helber @ahelber.bsky.social · 30/06/2026
I skimmed the article, i think it doesnt say; which OR solvers did you use? For staff scheduling many (commercial as well as free) tools exist. It would be interesting to know if you had to build something custom (and why) or if the challenge was more the integration into the system
130
Alexander Helber @ahelber.bsky.social · 23/06/2026
and oranges achievable together) can be helpful because then you get a feeling for the available tradeoffs. Finding all these non-dominated pairs is very hard for many problems. I suppose in games its often a bit easier because essentially the choice at any point is only "one apple" or "one orange'
000
Alexander Helber @ahelber.bsky.social · 23/06/2026
From our perspective, the response to unclear preferences between trade-offs is to determine the pareto-front of non-dominated moves, i.e. those that give you for example a fixed amount of apples and the highest amount of oranges possible then. Finding the front (i.e. all pairs of numbers of apples
100
Alexander Helber @ahelber.bsky.social · 23/06/2026
prefer an action simply because it aligns with their view of themselves or because they find it funny. Or even if you look only at actually winning, often "maximize placement" and "maximize points" are not the same and players might have a different notion of what "doing as best as you can" is
100
Alexander Helber @ahelber.bsky.social · 23/06/2026
determine the relative value of resources (which in some games can vary widely in usefulness over time). I'd argue that sometimes it is even possible that there is no way of determining a comparable utility value. A classic example might be (as has been raised in this thread) that some players might
100
Alexander Helber @ahelber.bsky.social · 23/06/2026
Also i would agree that most of your taxonomy concerns uncertainty/ambiguity, which as you explain is where the interest comes from. The one exception seems to be "apples versus oranges", which reminds me of what we would call multi-objective optimization. You mention that a key difficulty is to
110
Alexander Helber @ahelber.bsky.social · 23/06/2026
I'd say Operations Research, which is essentially meaningless to people not in at least related fields. Here you can read about the renaming proposal (which didnt go anywhere i would say) and a bit of the history pubsonline.informs.org/do/10.1287/o... The survey results are funny.
pubsonline.informs.org
Distinguishing the Profession of Operations Research in the Age of Analytics, Big Data, Data Science and AI | ORMS Today
What’s in a name? A question famously asked in the play, “Romeo and Juliet,” Shakespeare’s point is that a name by itself has no value or meaning, but rather is just a convention for distinguishing ob...
000
Alexander Helber @ahelber.bsky.social · 22/06/2026
information (about the game state or your intentions). I suppose in the end there are many ways to classify anything and the appropriate framework depends on the question you are trying to answer :)
100
Alexander Helber @ahelber.bsky.social · 22/06/2026
are distinguished by available information level and how many additional decisions you will be able to make to adjust the state after you make a decision now). Or physical versus financial versus informational decisions. These you also see in board games: change the board state, take a loan, reveal
100
Alexander Helber @ahelber.bsky.social · 22/06/2026
It is my last turn and i just have to do the optimal sequence of actions to maximize points) are usually a slog. Though sometimes it is fun to figure out the best way to run an engine or find a combo. Other classifications of decision types i have seen: strategic/tactical/operational (usually these
100
Alexander Helber @ahelber.bsky.social · 22/06/2026
A guaranteed best decision) and settings under uncertainty, where you can only hope to make decisions that are, for example, good on average. I suppose applied to games i prefer decisions under uncertainty (where i try to estimate likelihoods, player behaviour etc). Deterministic decisions (e.g.
100
Alexander Helber @ahelber.bsky.social · 22/06/2026
Interesting! Coming from a field that some are trying to rename "Decision Science", which has its own taxonomies of decisions, it is fascinating to see how a similar question plays out from a very different angle. A main distinction is perhaps between a deterministic setting (where you can determine
100
Alexander Helber @ahelber.bsky.social · 06/06/2026
I would have agreed until a few days ago, then i saw this: seems like for some use cases the detectors can kind of work. My assumption is its mostly the type of writing where you kind of already know that you are reading slop. And probably works better for long papers. For school essays? Nah
blog.neurips.cc
AI-Generated Papers in the NeurIPS 2026 Position Paper Track – NeurIPS Blog
000
Reposted by Alexander Helber
Mark Riedl @markriedl.bsky.social · 02/06/2026
NeurIPS statement on the use of AI in the position paper track blog.neurips.cc/2026/06/02/a...
blog.neurips.cc
AI-Generated Papers in the NeurIPS 2026 Position Paper Track – NeurIPS Blog
0127
Reposted by Alexander Helber
Thiago Serra @thserra.bsky.social · 18/05/2026
Sally Dong talked at the #mip2026 workshop about solving linear programs more efficiently by identifying underlying graphs associated with their constraint matrix. #orms 1/N
162
Reposted by Alexander Helber
Dr. Casey Fiesler @cfiesler.bsky.social · 16/05/2026
So one consistent thread is: we have no agency over our own behavior in the presence of temptation (zoning off during lectures, checking a phone in class, using gen AI), and that's your problem to solve. I know that structural interventions that reduce temptation are legitimate pedagogy but ugh.
3566
Alexander Helber @ahelber.bsky.social · 14/05/2026
Will there also be naming and shaming? Seems like introducing some reputation management aspects might increase the stakes a bit and introduce real deterrence ;)
180
Reposted by Alexander Helber
Victoria Rose Richards @aerialembroidery.bsky.social · 03/05/2026
'Beltsville Agricultural Research Centre', Maryland - I was privately commissioned to memorialise the BARC by a past employee after learning it will be shut down by the USDA after over 100 years of research. They wanted to focus of this aerial embroidery to be the field plots... 1/3 (see thread)
An embroidery hoop containing an aerial view embroidered landscape. It depicts the BARC in Maryland with green and brown field plots and a building area in the bottom right corner.A zoomed-in angled view of an aerial embroidery. It shows field plots and buildings.A zoomed-in angled view of an aerial embroidery. It shows field plots and buildings.
341793575
Alexander Helber @ahelber.bsky.social · 24/04/2026
Short paper available here: openproceedings.org/2026/conf/in... Note that the openproceedings website lists a third co-author (that we dont know), we are in contact with the organizers to get this person removed 😅
openproceedings.org
010
Alexander Helber @ahelber.bsky.social · 24/04/2026
We show that finding the smallest possible time discretization that prevents all illegal consolidations that have been observed so far is NP-hard. We propose a MIP formulation which does not solve well but demonstrates that if we could solve this problem fast, it would help the overall DDD algorithm
110
Alexander Helber @ahelber.bsky.social · 24/04/2026
Yesterday, Tom presented our work on better time discretization refinement for dynamic discretization discovery algorithms at INOC (held at HEC Liege). 1/3
Tom in front of the final takeaway slide: optimal refinement is NP hard, beneficial, but too slow with a MIP solverTom in front of a slide showing an example of the continuous time service network design  problem
121
Alexander Helber @ahelber.bsky.social · 10/04/2026
On another not, at least for computational fields it seems like AI may lower the (perceived) value of work on more established topics, which may give incentive to then spend more time on new lines of work.
010
Alexander Helber @ahelber.bsky.social · 10/04/2026
I (phd student) usually (try to) think about new ideas/topics whenever i dont want to work for whatever reason on the "established topics" :) Procrastination fueled by shiny ideas? Perhaps as a group lead it is harder to get "distracted" by random ideas in the daily stream of ongoing work.
210
Reposted by Alexander Helber
Minas Karamanis @minaskar.bsky.social · 30/03/2026
Hey, I wrote a thing about AI in astrophysics ergosphere.blog/posts/the-ma...
ergosphere.blog
The machines are fine. I'm worried about us.
On AI agents, grunt work, and the part of science that isn't replaceable.
1101781530
Reposted by Alexander Helber
Anna Nagurney @annanagurney.bsky.social · 02/04/2026
We have always had oral exams as part of our OM/MS PhD program @isenbergumass.bsky.social, with an oral exam being an important component of the core exam, the comprehensive exam, and also the dissertation proposal defense, plus the dissertation defense! apnews.com/article/coll...
apnews.com
Perfect homework, blank stares: Why colleges are turning to oral exams to combat AI
A growing number of U.S. college instructors are turning to oral exams to help combat an AI crisis in higher education.
033
Reposted by Alexander Helber
Marco Lübbecke @mluebbecke.bsky.social · 27/03/2026
13 PhD positions in in applied #maths, #orms, engineering, or CS within the ALMOA consortium -> study applications of European and international relevance in areas such as sustainable energy systems, green gogistics, etc. almoa.aau.at?page_id=53
almoa.aau.at
DC Projects – ALMOA
054
Alexander Helber @ahelber.bsky.social · 09/03/2026
It turns out, trains are very complicated and need simplification.. Today i had a look into a >300 page book on railway optimization and the idea of a train having to accelerate was only mentioned 10 times, none of them in the sections about operational optimization (so closest to "real physics"). 🚂
000
Alexander Helber @ahelber.bsky.social · 25/02/2026
Stephan Bogs of the Chair of Operations Management presented three case-studies that build on each other to teach students modelling sustainable supply chains, with multiple objectives and in stochastic settings. Also openly available here: www.orca.nrw/oer/oer-find...
020
Alexander Helber @ahelber.bsky.social · 25/02/2026
Matthias Gehnen of the Theoretical Computer Science Group presented the "Algorithmic Battle", in which groups are pitched against each other: for a given problem class, you have to write a solver and an instance generator, then you try to solve each others instances. Website available open-source!
A chart showing three teams score over time
000
Alexander Helber @ahelber.bsky.social · 25/02/2026
Tim Donkiewicz of the Chair of Operations Research presented the OR Practice Project, in which teams of students tackle real-world problems brought in by practitioners. The course demands a lot of students, so an important point was the introduction of a Self-Assessment to communicate expectations.
000
Alexander Helber @ahelber.bsky.social · 25/02/2026
Yesterday we organized a presentation session about teaching practical hands-on courses for the Center of Algorithmics and Optimization (algoopt.rwth-aachen.de) at RWTH Aachen.
300
Alexander Helber @ahelber.bsky.social · 20/02/2026
It seems we will do that plus benders and won't implement cutting plane separation methods with the students.
010
Alexander Helber @ahelber.bsky.social · 20/02/2026
That is also what we arrived at - TSP with DFJ SECs is what worked in the past and shall also work this year. @mluebbecke.bsky.social observed that it has the added benefit of showing students more tools for modelling (i.e., handling an exponential number of constraints).
120
Alexander Helber @ahelber.bsky.social · 18/02/2026
So this may be a good example of better formulations without necessarily separating cuts
100
Alexander Helber @ahelber.bsky.social · 18/02/2026
Upon further investigation, it seems (at least for the generated instances) that the cutting plane separation is not helpful, especially for the extended formulation, which usually finishes in the root node anyway and where cuts are almost never separated.
100
Alexander Helber @ahelber.bsky.social · 18/02/2026
So i shall read in more detail and see if we can present these formulations on a level appropriate for this course. Thanks for the suggestion!
000