Tootfinder

Cactus flower spaces and monodromy of Bethe vectors
Joel Kamnitzer, Leonid Rybnikov
https://arxiv.org/abs/2507.12829 https://arxiv.or…

Cactus flower spaces and monodromy of Bethe vectors
We continue the study of cactus flower moduli spaces $\overline{F}_n$ and Gaudin models started in arXiv:2308.06880, arXiv:2407.06424. We show that isomorphism classes of operadic coverings of the real form $\overline{F}_n(\mathbb{R})$ are naturally one-to-one with equivalence classes of concrete coboundary monoidal categories (i.e. coboundary monoidal categories that admit a faithful monoidal functor to sets) with certain semisimplicity and finiteness conditions. Following the strategy of arXi…

AI as a teaching tool and learning partner
Steven Watterson, Sarah Atkinson, Elaine Murray, Andrew McDowell
https://arxiv.org/abs/2509.13899 https://arxiv.…

AI as a teaching tool and learning partner
The arrival of AI tools and in particular Large Language Models (LLMs) has had a transformative impact on teaching and learning and institutes are still trying to determine how to integrate LLMs into education in constructive ways. Here, we explore the adoption of LLM-based tools into two teaching programmes, one undergraduate and one postgraduate. We provided to our classes (1) a LLM-powered chatbot that had access to course materials by RAG and (2) AI-generated audio-only podcasts for each we…

Should we teach vibe coding? Here's why not.
Should AI coding be taught in undergrad CS education?
1/2
I teach undergraduate computer science labs, including for intro and more-advanced core courses. I don't publish (non-negligible) scholarly work in the area, but I've got years of craft expertise in course design, and I do follow the academic literature to some degree. In other words, In not the world's leading expert, but I have spent a lot of time thinking about course design, and consider myself competent at it, with plenty of direct experience in what knowledge & skills I can expect from students as they move through the curriculum.
I'm also strongly against most uses of what's called "AI" these days (specifically, generative deep neutral networks as supplied by our current cadre of techbro). There are a surprising number of completely orthogonal reasons to oppose the use of these systems, and a very limited number of reasonable exceptions (overcoming accessibility barriers is an example). On the grounds of environmental and digital-commons-pollution costs alone, using specifically the largest/newest models is unethical in most cases.
But as any good teacher should, I constantly question these evaluations, because I worry about the impact on my students should I eschew teaching relevant tech for bad reasons (and even for his reasons). I also want to make my reasoning clear to students, who should absolutely question me on this. That inspired me to ask a simple question: ignoring for one moment the ethical objections (which we shouldn't, of course; they're very stark), at what level in the CS major could I expect to teach a course about programming with AI assistance, and expect students to succeed at a more technically demanding final project than a course at the same level where students were banned from using AI? In other words, at what level would I expect students to actually benefit from AI coding "assistance?"
To be clear, I'm assuming that students aren't using AI in other aspects of coursework: the topic of using AI to "help you study" is a separate one (TL;DR it's gross value is not negative, but it's mostly not worth the harm to your metacognitive abilities, which AI-induced changes to the digital commons are making more important than ever).
So what's my answer to this question?
If I'm being incredibly optimistic, senior year. Slightly less optimistic, second year of a masters program. Realistic? Maybe never.
The interesting bit for you-the-reader is: why is this my answer? (Especially given that students would probably self-report significant gains at lower levels.) To start with, [this paper where experienced developers thought that AI assistance sped up their work on real tasks when in fact it slowed it down] (https://arxiv.org/abs/2507.09089) is informative. There are a lot of differences in task between experienced devs solving real bugs and students working on a class project, but it's important to understand that we shouldn't have a baseline expectation that AI coding "assistants" will speed things up in the best of circumstances, and we shouldn't trust self-reports of productivity (or the AI hype machine in general).
Now we might imagine that coding assistants will be better at helping with a student project than at helping with fixing bugs in open-source software, since it's a much easier task. For many programming assignments that have a fixed answer, we know that many AI assistants can just spit out a solution based on prompting them with the problem description (there's another elephant in the room here to do with learning outcomes regardless of project success, but we'll ignore this over too, my focus here is on project complexity reach, not learning outcomes). My question is about more open-ended projects, not assignments with an expected answer. Here's a second study (by one of my colleagues) about novices using AI assistance for programming tasks. It showcases how difficult it is to use AI tools well, and some of these stumbling blocks that novices in particular face.
But what about intermediate students? Might there be some level where the AI is helpful because the task is still relatively simple and the students are good enough to handle it? The problem with this is that as task complexity increases, so does the likelihood of the AI generating (or copying) code that uses more complex constructs which a student doesn't understand. Let's say I have second year students writing interactive websites with JavaScript. Without a lot of care that those students don't know how to deploy, the AI is likely to suggest code that depends on several different frameworks, from React to JQuery, without actually setting up or including those frameworks, and of course three students would be way out of their depth trying to do that. This is a general problem: each programming class carefully limits the specific code frameworks and constructs it expects students to know based on the material it covers. There is no feasible way to limit an AI assistant to a fixed set of constructs or frameworks, using current designs. There are alternate designs where this would be possible (like AI search through adaptation from a controlled library of snippets) but those would be entirely different tools.
So what happens on a sizeable class project where the AI has dropped in buggy code, especially if it uses code constructs the students don't understand? Best case, they understand that they don't understand and re-prompt, or ask for help from an instructor or TA quickly who helps them get rid of the stuff they don't understand and re-prompt or manually add stuff they do. Average case: they waste several hours and/or sweep the bugs partly under the rug, resulting in a project with significant defects. Students in their second and even third years of a CS major still have a lot to learn about debugging, and usually have significant gaps in their knowledge of even their most comfortable programming language. I do think regardless of AI we as teachers need to get better at teaching debugging skills, but the knowledge gaps are inevitable because there's just too much to know. In Python, for example, the LLM is going to spit out yields, async functions, try/finally, maybe even something like a while/else, or with recent training data, the walrus operator. I can't expect even a fraction of 3rd year students who have worked with Python since their first year to know about all these things, and based on how students approach projects where they have studied all the relevant constructs but have forgotten some, I'm not optimistic seeing these things will magically become learning opportunities. Student projects are better off working with a limited subset of full programming languages that the students have actually learned, and using AI coding assistants as currently designed makes this impossible. Beyond that, even when the "assistant" just introduces bugs using syntax the students understand, even through their 4th year many students struggle to understand the operation of moderately complex code they've written themselves, let alone written by someone else. Having access to an AI that will confidently offer incorrect explanations for bugs will make this worse.
To be sure a small minority of students will be able to overcome these problems, but that minority is the group that has a good grasp of the fundamentals and has broadened their knowledge through self-study, which earlier AI-reliant classes would make less likely to happen. In any case, I care about the average student, since we already have plenty of stuff about our institutions that makes life easier for a favored few while being worse for the average student (note that our construction of that favored few as the "good" students is a large part of this problem).
To summarize: because AI assistants introduce excess code complexity and difficult-to-debug bugs, they'll slow down rather than speed up project progress for the average student on moderately complex projects. On a fixed deadline, they'll result in worse projects, or necessitate less ambitious project scoping to ensure adequate completion, and I expect this remains broadly true through 4-6 years of study in most programs (don't take this as an endorsement of AI "assistants" for masters students; we've ignored a lot of other problems along the way).
There's a related problem: solving open-ended project assignments well ultimately depends on deeply understanding the problem, and AI "assistants" allow students to put a lot of code in their file without spending much time thinking about the problem or building an understanding of it. This is awful for learning outcomes, but also bad for project success. Getting students to see the value of thinking deeply about a problem is a thorny pedagogical puzzle at the best of times, and allowing the use of AI "assistants" makes the problem much much worse. This is another area I hope to see (or even drive) pedagogical improvement in, for what it's worth.
1/2

Addressing Data Imbalance in Transformer-Based Multi-Label Emotion Detection with Weighted Loss
Xia Cui
https://arxiv.org/abs/2507.11384 https://

Addressing Data Imbalance in Transformer-Based Multi-Label Emotion Detection with Weighted Loss
This paper explores the application of a simple weighted loss function to Transformer-based models for multi-label emotion detection in SemEval-2025 Shared Task 11. Our approach addresses data imbalance by dynamically adjusting class weights, thereby enhancing performance on minority emotion classes without the computational burden of traditional resampling methods. We evaluate BERT, RoBERTa, and BART on the BRIGHTER dataset, using evaluation metrics such as Micro F1, Macro F1, ROC-AUC, Accurac…

Using Integer Programming to Solve Games, Puzzles, and Ciphers
Elizabeth Bouzarth, John Harris, Kevin Hutson, Christian Millichap
https://arxiv.org/abs/2509.12174 https://

Using Integer Programming to Solve Games, Puzzles, and Ciphers
In this paper, we introduce three different classes of undergraduate research projects that implement model building and integer programming. These research projects focus on determining and analyzing solutions to the game The Genius Square, optimizing allocation of trains to maximize points in the game Ticket to Ride, and (code)breaking monoalphabetic substitution ciphers. Initial models and analyses for these scenarios that came from previous undergraduate research projects are shared along w…

The $\alpha$-Cap Process: A Continuous Model for Random Geometric Networks of Binary Neurons
Mirabel Reid, Daniel J. Zhang
https://arxiv.org/abs/2508.09396 https://

The $α$-Cap Process: A Continuous Model for Random Geometric Networks of Binary Neurons
Recurrent networks of binary neurons are a foundational concept in artificial intelligence. While these networks are traditionally assumed to be fully connected, complex dynamics can emerge when the graph structure is varied. One graph structure of particular interest is the geometric random graph, which models the spatial dependencies present in biological neural networks. In such classes of graphs, global state dependencies tend to complicate analysis, motivating the study of their dynamics i…

Short-term CO2 emissions forecasting: insight from the Italian electricity market
Pierdomenico Duttilo, Francesco Lisi
https://arxiv.org/abs/2507.12992 htt…

Short-term CO2 emissions forecasting: insight from the Italian electricity market
This study investigates the short-term forecasting of carbon emissions from electricity generation in the Italian power market. Using hourly data from 2021 to 2023, several statistical models and forecast combination methods are evaluated and compared at the national and zonal levels. Four main model classes are considered: (i) linear parametric models, such as seasonal autoregressive integrated moving average and its exogenousvariable extension; (ii) functional parametric models, including sea…

Testing Viability of Benchmark Dark Matter Models for the Galactic Center Excess
Yongao Hu, Cari Cesarotti, Tracy R. Slatyer
https://arxiv.org/abs/2509.08043 https://

Testing Viability of Benchmark Dark Matter Models for the Galactic Center Excess
We examine the status of benchmark simplified dark matter models that have been proposed to explain the GeV gamma-ray Galactic Center excess. We constrain the available parameter space using updated observations from indirect detection, cosmology, direct detection, and accelerators. We show that there is still unconstrained parameter space in both classes of models we consider (a secluded dark sector with a vector portal coupling, and a two-Higgs doublet with a pseudoscalar mediator), and discu…

Should We Always Train Models on Fine-Grained Classes?
Davide Pirovano, Federico Milanesio, Michele Caselle, Piero Fariselli, Matteo Osella
https://arxiv.org/abs/2509.05130 http…

Should We Always Train Models on Fine-Grained Classes?
In classification problems, models must predict a class label based on the input data features. However, class labels are organized hierarchically in many datasets. While a classification task is often defined at a specific level of this hierarchy, training can utilize a finer granularity of labels. Empirical evidence suggests that such fine-grained training can enhance performance. In this work, we investigate the generality of this observation and explore its underlying causes using both real…

VGGSounder: Audio-Visual Evaluations for Foundation Models
Daniil Zverev, Thadd\"aus Wiedemer, Ameya Prabhu, Matthias Bethge, Wieland Brendel, A. Sophia Koepke
https://arxiv.org/abs/2508.08237

VGGSounder: Audio-Visual Evaluations for Foundation Models
The emergence of audio-visual foundation models underscores the importance of reliably assessing their multi-modal understanding. The VGGSounder dataset is commonly used as a benchmark for evaluation audio-visual classification. However, our analysis identifies several limitations of VGGSounder, including incomplete labelling, partially overlapping classes, and misaligned modalities. These lead to distorted evaluations of auditory and visual capabilities. To address these limitations, we introd…

FactorHD: A Hyperdimensional Computing Model for Multi-Object Multi-Class Representation and Factorization
Yifei Zhou, Xuchu Huang, Chenyu Ni, Min Zhou, Zheyu Yan, Xunzhao Yin, Cheng Zhuo
https://arxiv.org/abs/2507.12366

FactorHD: A Hyperdimensional Computing Model for Multi-Object Multi-Class Representation and Factorization
Neuro-symbolic artificial intelligence (neuro-symbolic AI) excels in logical analysis and reasoning. Hyperdimensional Computing (HDC), a promising brain-inspired computational model, is integral to neuro-symbolic AI. Various HDC models have been proposed to represent class-instance and class-class relations, but when representing the more complex class-subclass relation, where multiple objects associate different levels of classes and subclasses, they face challenges for factorization, a crucia…

Cosmology in warped massive gravity
Sebastian Garcia-Saenz, Yuxiang Wei, Xue Zhou
https://arxiv.org/abs/2509.09270 https://arxiv.org/pdf/2509.09270

Cosmology in warped massive gravity
We study the cosmological dynamics and predictions in the theory of warped massive gravity. This set-up postulates a five-dimensional ghost-free massive graviton with a brane-localized four-dimensional massive gravity potential, and has the virtue of raising the strong-coupling scale of the 4D theory. We identify two classes of models that lead to decoupled equations for the scale factor on the brane: one characterized by a particular choice of boundary conditions for the Stückelberg fields an…

Why AI can't possibly make you more productive; long
#AI and "productivity", some thoughts:
Edit: fixed some typos.
Productivity is a concept that isn't entirely meaningless outside the context of capitalism, but it's a concept that is heavily inflected in a capitalist context. In many uses today it effectively means "how much you can satisfy and/or exceed your boss' expectations." This is not really what it should mean: even in an anarchist utopia, people would care about things like how many shirts they can produce in a week, although in an "I'd like to voluntarily help more people" way rather than an "I need to meet this quota to earn my survival" way. But let's roll with this definition for a second, because it's almost certainly what your boss means when they say "productivity", and understanding that word in a different (even if truer) sense is therefore inherently dangerous.
Accepting "productivity" to mean "satisfying your boss' expectations," I will now claim: the use of generative AI cannot increase your productivity.
Before I dive in, it's imperative to note that the big generative models which most people think of as constituting "AI" today are evil. They are 1: pouring fuel on our burning planet, 2: psychologically strip-mining a class of data laborers who are exploited for their precarity, 3: enclosing, exploiting, and polluting the digital commons, and 4: stealing labor from broad classes of people many of whom are otherwise glad to give that labor away for free provided they get a simple acknowledgement in return. Any of these four "ethical issues" should be enough *alone* to cause everyone to simply not use the technology. These ethical issues are the reason that I do not use generative AI right now, except for in extremely extenuating circumstances. These issues are also convincing for a wide range of people I talk to, from experts to those with no computer science background. So before I launch into a critique of the effectiveness of generative AI, I want to emphasize that such a critique should be entirely unnecessary.
But back to my thesis: generative AI cannot increase your productivity, where "productivity" has been defined as "how much you can satisfy and/or exceed your boss' expectations."
Why? In fact, what the fuck? Every AI booster I've met has claimed the opposite. They've given me personal examples of time saved by using generative AI. Some of them even truly believe this. Sometimes I even believe they saved time without horribly compromising on quality (and often, your boss doesn't care about quality anyways if the lack of quality is hard to measure of doesn't seem likely to impact short-term sales/feedback/revenue). So if generative AI genuinely lets you write more emails in a shorter period of time, or close more tickets, or something else along these lines, how can I say it isn't increasing your ability to meet your boss' expectations?
The problem is simple: your boss' expectations are not a fixed target. Never have been. In virtue of being someone who oversees and pays wages to others under capitalism, your boss' game has always been: pay you less than the worth of your labor, so that they can accumulate profit and thus more capital to remain in charge instead of being forced into working for a wage themselves. Sure, there are layers of management caught in between who aren't fully in this mode, but they are irrelevant to this analysis. It matters not how much you please your manager if your CEO thinks your work is not worth the wages you are being paid. And using AI actively lowers the value of your work relative to your wages.
Why do I say that? It's actually true in several ways. The most obvious: using generative AI lowers the quality of your work, because the work it produces is shot through with errors, and when your job is reduced to proofreading slop, you are bound to tire a bit, relax your diligence, and let some mistakes through. More than you would have if you are actually doing and taking pride in the work. Examples are innumerable and frequent, from journalists to lawyers to programmers, and we laugh at them "haha how stupid to not check whether the books the AI reviewed for you actually existed!" but on a deeper level if we're honest we know we'd eventually make the same mistake ourselves (bonus game: spot the swipe-typing typos I missed in this post; I'm sure there will be some).
But using generative AI also lowers the value of your work in another much more frightening way: in this era of hype, it demonstrates to your boss that you could be replaced by AI. The more you use it, and no matter how much you can see that your human skills are really necessary to correct its mistakes, the more it appears to your boss that they should hire the AI instead of you. Or perhaps retain 10% of the people in roles like yours to manage the AI doing the other 90% of the work. Paradoxically, the *more* you get done in terms of raw output using generative AI, the more it looks to your boss as if there's an opportunity to get enough work done with even fewer expensive humans. Of course, the decision to fire you and lean more heavily into AI isn't really a good one for long-term profits and success, but the modern boss did not get where they are by considering long-term profits. By using AI, you are merely demonstrating your redundancy, and the more you get done with it, the more redundant you seem.
In fact, there's even a third dimension to this: by using generative AI, you're also providing its purveyors with invaluable training data that allows them to make it better at replacing you. It's generally quite shitty right now, but the more use it gets by competent & clever people, the better it can become at the tasks those specific people use it for. Using the currently-popular algorithm family, there are limits to this; I'm not saying it will eventually transcend the mediocrity it's entwined with. But it can absolutely go from underwhelmingly mediocre to almost-reasonably mediocre with the right training data, and data from prompting sessions is both rarer and more useful than the base datasets it's built on.
For all of these reasons, using generative AI in your job is a mistake that will likely lead to your future unemployment. To reiterate, you should already not be using it because it is evil and causes specific and inexcusable harms, but in case like so many you just don't care about those harms, I've just explained to you why for entirely selfish reasons you should not use it.
If you're in a position where your boss is forcing you to use it, my condolences. I suggest leaning into its failures instead of trying to get the most out of it, and as much as possible, showing your boss very clearly how it wastes your time and makes things slower. Also, point out the dangers of legal liability for its mistakes, and make sure your boss is aware of the degree to which any of your AI-eager coworkers are producing low-quality work that harms organizational goals.
Also, if you've read this far and aren't yet of an anarchist mindset, I encourage you to think about the implications of firing 75% of (at least the white-collar) workforce in order to make more profit while fueling the climate crisis and in most cases also propping up dictatorial figureheads in government. When *either* the AI bubble bursts *or* if the techbros get to live out the beginnings of their worker-replacement fantasies, there are going to be an unimaginable number of economically desperate people living in increasingly expensive times. I'm the kind of optimist who thinks that the resulting social crucible, though perhaps through terrible violence, will lead to deep social changes that effectively unseat from power the ultra-rich that continue to drag us all down this destructive path, and I think its worth some thinking now about what you might want the succeeding stable social configuration to look like so you can advocate towards that during points of malleability.
As others have said more eloquently, generative AI *should* be a technology that makes human lives on average easier, and it would be were it developed & controlled by humanists. The only reason that it's not, is that it's developed and controlled by terrible greedy people who use their unfairly hoarded wealth to immiserate the rest of us in order to maintain their dominance. In the long run, for our very survival, we need to depose them, and I look forward to what the term "generative AI" will mean after that finally happens.

Dataset Ownership in the Era of Large Language Models
Kun Li, Cheng Wang, Minghui Xu, Yue Zhang, Xiuzhen Cheng
https://arxiv.org/abs/2509.05921 https://arx…

Dataset Ownership in the Era of Large Language Models
As datasets become critical assets in modern machine learning systems, ensuring robust copyright protection has emerged as an urgent challenge. Traditional legal mechanisms often fail to address the technical complexities of digital data replication and unauthorized use, particularly in opaque or decentralized environments. This survey provides a comprehensive review of technical approaches for dataset copyright protection, systematically categorizing them into three main classes: non-intrusive…

Generalized Zero-Shot Learning for Point Cloud Segmentation with Evidence-Based Dynamic Calibration
Hyeonseok Kim, Byeongkeun Kang, Yeejin Lee
https://arxiv.org/abs/2509.08280 h…

Generalized Zero-Shot Learning for Point Cloud Segmentation with Evidence-Based Dynamic Calibration
Generalized zero-shot semantic segmentation of 3D point clouds aims to classify each point into both seen and unseen classes. A significant challenge with these models is their tendency to make biased predictions, often favoring the classes encountered during training. This problem is more pronounced in 3D applications, where the scale of the training data is typically smaller than in image-based tasks. To address this problem, we propose a novel method called E3DPC-GZSL, which reduces overconf…

Modeling Non-Gaussianities in Pulsar Timing Array data analysis using Gaussian Mixture Models
Mikel Falxa, Alberto Sesana
https://arxiv.org/abs/2508.08365 https://

Modeling Non-Gaussianities in Pulsar Timing Array data analysis using Gaussian Mixture Models
In Pulsar Timing Array (PTA) data analysis, noise is typically assumed to be Gaussian, and the marginalized likelihood has a well-established analytical form derived within the framework of Gaussian processes. However, this Gaussianity assumption may break down for certain classes of astrophysical and cosmological signals, particularly for a gravitational wave background (GWB) generated by a population of supermassive black hole binaries (SMBHBs). In this work, we present a new method for testi…

Analyzing Finnish Inflectional Classes through Discriminative Lexicon and Deep Learning Models
Alexandre Nikolaev, Yu-Ying Chuang, R. Harald Baayen
https://arxiv.org/abs/2509.04813

Analyzing Finnish Inflectional Classes through Discriminative Lexicon and Deep Learning Models
Descriptions of complex nominal or verbal systems make use of inflectional classes. Inflectional classes bring together nouns which have similar stem changes and use similar exponents in their paradigms. Although inflectional classes can be very useful for language teaching as well as for setting up finite state morphological systems, it is unclear whether inflectional classes are cognitively real, in the sense that native speakers would need to discover these classes in order to learn how to p…

Projection Cascades of models of log del Pezzo surfaces
Muhammad Imran Qureshi
https://arxiv.org/abs/2506.19970 https://arxiv.org/pdf…

Projection Cascades of models of log del Pezzo surfaces
We introduce the notion of type-I projection cascade for a biregular model (infinite series) of log del Pezzo surfaces. We study the existence of type-I projection cascades for known classes of models of log del Pezzo surfaces, such that their images under their anti-canonical embeddings in some weighted projective space, can be described as codimension 4 and codimension 3 varieties. We obtain two cascades of length three and four cascades of length two, where each projection gives rise to a we…

Mathematical modelling in Physics: deterministic processes
Sergej Pankratow
https://arxiv.org/abs/2507.08004 https://arxiv.org/pdf/25…

Mathematical modelling in Physics: deterministic processes
Basic principles of mathematical modeling are reviewed in this book, with the focus on physics and its practical applications, and examples of selected mathematical methods are presented. Most of the models have been imported from physics and applied for scientific and engineering problems. More specifically, the accent is placed on the theory of dynamical systems being used to produce evolutionary models. Classes of different models are presented and the relationship of mathematical modeling t…

Governing AI R&D: A Legal Framework for Constraining Dangerous AI
Alex Mark, Aaron Scher
https://arxiv.org/abs/2509.05361 https://arxiv.org/pdf/2509.05…

Governing AI R&D: A Legal Framework for Constraining Dangerous AI
As AI advances, governing its development may become paramount to public safety. Lawmakers may seek to restrict the development and release of AI models or of AI research itself. These governance actions could trigger legal challenges that invalidate the actions, so lawmakers should consider these challenges ahead of time. We investigate three classes of potential litigation risk for AI regulation in the U.S.: the First Amendment, administrative law, and the Fourteenth Amendment. We discuss exi…

New universality classes govern the critical and multicritical behavior of an active Ising model
Matthew Wong, Chiu Fan Lee
https://arxiv.org/abs/2507.06068

New universality classes govern the critical and multicritical behavior of an active Ising model
The Ising model is one of the most well known models in statistical physics, with its critical behavior governed by the Wilson-Fisher universality class (UC). When active motility is incorporated into the Ising model by, e.g., dictating that the spins' directional movements follow their orientations, the spin number density necessarily constitutes a soft mode in the hydrodynamic description, and can therefore modify the scaling behavior of the system. Here, we show that this is indeed the case …

The Landscape of Collapsar Outflows: Structure, Signatures and Origins of Einstein Probe Relativistic Supernova Transients
Ore Gottlieb
https://arxiv.org/abs/2509.04551 https://…

The Landscape of Collapsar Outflows: Structure, Signatures and Origins of Einstein Probe Relativistic Supernova Transients
The Einstein Probe is revolutionizing time-domain astrophysics through the discovery of new classes of X-ray transients associated with Type Ic-Broad Line supernovae. These events commonly exhibit bright early-time optical counterparts, and sudden afterglow rebrightening within the first week - features that existing models fail to explain. In particular, structured jet and cocoon scenarios are inconsistent with the observed sharp rebrightening and multi-day optical emission, while the refreshe…

Path integral approach to quantum thermalization
Alexander Altland, Kun Woo Kim, Tobias Micklitz
https://arxiv.org/abs/2509.06028 https://arxiv.org/pdf/250…

Path integral approach to quantum thermalization
We introduce a quasiclassical Green function approach describing the unitary yet irreversible dynamics of quantum systems effectively acting as their own environment. Combining a variety of concepts of quantum many-body theory, notably the nonlinear $σ$-model of disordered systems, the $G Σ$-formalism for strong correlations, and real time path integration, the theory is capable of describing a wide range of system classes and disorder models. It extends previous work beyond perturbation theo…

Hexagonal Picture Scanning Automata
Deepalakshmi D, Lisa Mathew
https://arxiv.org/abs/2508.07779 https://arxiv.org/pdf/2508.07779

Hexagonal Picture Scanning Automata
Two new classes of finite automata, called General hexagonal Boustrophedon finite automata and General hexagonal returning finite automata operating on hexagonal grids, are introduced and analyzed. The work establishes the theoretical foundations for these automata models, examines their computational properties, and investigates the relationships and equivalences between the language families they define. The research contributes to the broader understanding of two-dimensional automata theory …

SA-OOSC: A Multimodal LLM-Distilled Semantic Communication Framework for Enhanced Coding Efficiency with Scenario Understanding
Feifan Zhang, Yuyang Du, Yifan Xiang, Xiaoyan Liu, Soung Chang Liew
https://arxiv.org/abs/2509.07436

SA-OOSC: A Multimodal LLM-Distilled Semantic Communication Framework for Enhanced Coding Efficiency with Scenario Understanding
This paper introduces SA-OOSC, a multimodal large language models (MLLM)-distilled semantic communication framework that achieves efficient semantic coding with scenario-aware importance allocations. This approach addresses a critical limitation of existing object-oriented semantic communication (OOSC) systems - assigning static importance values to specific classes of objects regardless of their contextual relevance. Our framework utilizes MLLMs to identify the scenario-augmented (SA) semantic…

Just read this post by @… on an optimistic AGI future, and while it had some interesting and worthwhile ideas, it's also in my opinion dangerously misguided, and plays into the current AGI hype in a harmful way.
https://social.coop/@eloquence/114940607434005478
My criticisms include:
- Current LLM technology has many layers, but the biggest most capable models are all tied to corporate datacenters and require inordinate amounts of every and water use to run. Trying to use these tools to bring about a post-scarcity economy will burn up the planet. We urgently need more-capable but also vastly more efficient AI technologies if we want to use AI for a post-scarcity economy, and we are *not* nearly on the verge of this despite what the big companies pushing LLMs want us to think.
- I can see that permacommons.org claims a small level of expenses on AI equates to low climate impact. However, given current deep subsidies on place by the big companies to attract users, that isn't a great assumption. The fact that their FAQ dodges the question about which AI systems they use isn't a great look.
- These systems are not free in the same way that Wikipedia or open-source software is. To run your own model you need a data harvesting & cleaning operation that costs millions of dollars minimum, and then you need millions of dollars worth of storage & compute to train & host the models. Right now, big corporations are trying to compete for market share by heavily subsidizing these things, but it you go along with that, you become dependent on them, and you'll be screwed when they jack up the price to a profitable level later. I'd love to see open dataset initiatives SBD the like, and there are some of these things, but not enough yet, and many of the initiatives focus on one problem while ignoring others (fine for research but not the basis for a society yet).
- Between the environmental impacts, the horrible labor conditions and undercompensation of data workers who filter the big datasets, and the impacts of both AI scrapers and AI commons pollution, the developers of the most popular & effective LLMs have a lot of answer for. This project only really mentions environmental impacts, which makes me think that they're not serious about ethics, which in turn makes me distrustful of the whole enterprise.
- Their language also ends up encouraging AI use broadly while totally ignoring several entire classes of harm, so they're effectively contributing to AI hype, especially with such casual talk of AGI and robotics as if embodied AGI were just around the corner. To be clear about this point: we are several breakthroughs away from AGI under the most optimistic assumptions, and giving the impression that those will happen soon plays directly into the hands of the Sam Altmans of the world who are trying to make money off the impression of impending huge advances in AI capabilities. Adding to the AI hype is irresponsible.
- I've got a more philosophical criticism that I'll post about separately.
I do think that the idea of using AI & other software tools, possibly along with robotics and funded by many local cooperatives, in order to make businesses obsolete before they can do the same to all workers, is a good one. Get your local library to buy a knitting machine alongside their 3D printer.
Lately I've felt too busy criticizing AI to really sit down and think about what I do want the future to look like, even though I'm a big proponent of positive visions for the future as a force multiplier for criticism, and this article is inspiring to me in that regard, even if the specific project doesn't seem like a good one.

Isoprobability Models of Qubit Dynamics: Demonstration via Time-Dependent Phase Control on IBM Quantum
Ivo S. Mihov, Nikolay V. Vitanov
https://arxiv.org/abs/2506.19572

Isoprobability Models of Qubit Dynamics: Demonstration via Time-Dependent Phase Control on IBM Quantum
Efficient quantum control is a cornerstone for the advancement of quantum technologies, from computation to sensing and communications. Several approaches in quantum control, e.g. optimal control and inverse engineering, use the pulse amplitude and frequency shaping as the control tools. Often, these approaches prescribe pulse shapes which are difficult or impossible to implement. To this end, we develop the concept of isoprobability classes of models of qubit dynamics, in which various pairs o…

Viability of generalized $\alpha$-inflation from Planck, ACT, and DESI Data
Gabriel German, Juan Carlos Hidalgo
https://arxiv.org/abs/2508.01017 https://ar…

Viability of generalized $α$-inflation from Planck, ACT, and DESI Data
We study two classes of single-field inflationary models - a generalization of the alpha-attractor and the alpha-Starobinsky model - and examine their compatibility with current observational data from Planck, ACT DR6, and BAO measurements from DESI DR2. Our analysis focuses on the reheating phase that follows inflation, characterized by the equation-of-state parameter omega_re, the reheating temperature T_re, and the number of e-folds N_re. We use a semi-analytical approach based on an equatio…

Toward Accessible Dermatology: Skin Lesion Classification Using Deep Learning Models on Mobile-Acquired Images
Asif Newaz, Masum Mushfiq Ishti, A Z M Ashraful Azam, Asif Ur Rahman Adib
https://arxiv.org/abs/2509.04800

Toward Accessible Dermatology: Skin Lesion Classification Using Deep Learning Models on Mobile-Acquired Images
Skin diseases are among the most prevalent health concerns worldwide, yet conventional diagnostic methods are often costly, complex, and unavailable in low-resource settings. Automated classification using deep learning has emerged as a promising alternative, but existing studies are mostly limited to dermoscopic datasets and a narrow range of disease classes. In this work, we curate a large dataset of over 50 skin disease categories captured with mobile devices, making it more representative o…

Collective contributions to polarization in political voting
Gavin Rees, Edward D. Lee
https://arxiv.org/abs/2508.02496 https://arxiv.org/pdf/2508.02496

Collective contributions to polarization in political voting
Politics around the world exhibits increasing polarization, demonstrated in part by rigid voting configurations in legislatures. The crux of polarization is separation along a unidimensional ideological axis, but how it emerges is yet partially understood. We refine a powerful class of models from statistical physics, restricted Boltzmann machines, to unify two classes of individual voter preference and voter interaction models. Our model is minimally parameterized, fits voting data well, and h…

Leveraging LLMs for Automated Translation of Legacy Code: A Case Study on PL/SQL to Java Transformation
Lola Solovyeva, Eduardo Carneiro Oliveira, Shiyu Fan, Alper Tuncay, Shamil Gareev, Andrea Capiluppi
https://arxiv.org/abs/2508.19663

Leveraging LLMs for Automated Translation of Legacy Code: A Case Study on PL/SQL to Java Transformation
The VT legacy system, comprising approximately 2.5 million lines of PL/SQL code, lacks consistent documentation and automated tests, posing significant challenges for refactoring and modernisation. This study investigates the feasibility of leveraging large language models (LLMs) to assist in translating PL/SQL code into Java for the modernised "VTF3" system. By leveraging a dataset comprising 10 PL/SQL-to-Java code pairs and 15 Java classes, which collectively established a domain model for th…

Wilson Line and Disorder Invariants of Topological One-Dimensional Multiband Models
R. Moola, A. Mckenna, M. Hilke
https://arxiv.org/abs/2507.01846 https:/…

Wilson Line and Disorder Invariants of Topological One-Dimensional Multiband Models
Topological invariants, such as the winding number, the Chern number, and the Zak phase, characterize the topological phases of bulk materials. Through the bulk-boundary correspondence, these topological phases have a one-to-one correspondence to topological edge states, which are robust to certain classes of disorder. For simple models like the Su-Schrieffer-Heeger (SSH) model, the computation of the winding number and Zak phase are straightforward, however, in multiband systems, this is no lo…

Statistical Inference for Subgraph Frequencies of Exchangeable Hyperedge Models
Ayoushman Bhattacharya, Nilanjan Chakraborty, Robert Lunde
https://arxiv.org/abs/2508.13258 https…

Statistical Inference for Subgraph Frequencies of Exchangeable Hyperedge Models
In statistical network analysis, models for binary adjacency matrices satisfying vertex exchangeability are commonly used. However, such models may fail to capture key features of the data-generating process when interactions, rather than nodes, are fundamental units. We study statistical inference for subgraph counts under an exchangeable hyperedge model. We introduce several classes of subgraph statistics for hypergraphs and develop inferential tools for subgraph frequencies that account for …

Geometric models of simple Lie algebras via singularity theory
Cheol-Hyun Cho, Wonbo Jeong, Beom-Seok Kim
https://arxiv.org/abs/2507.22836 https://arxiv.or…

Geometric models of simple Lie algebras via singularity theory
It is well-known that ADE Dynkin diagrams classify both the simply-laced simple Lie algebras and simple singularities. We introduce a polygonal wheel in a plane for each case of ADE, called the Coxeter wheel. We show that equivalence classes of edges and spokes of a Coxeter wheel form a geometric root system isomorphic to the classical root system of the corresponding type. This wheel is in fact derived from the Milnor fiber of corresponding simple singularities of two variables, and the biline…

Generalized Quantum Hadamard Test for Machine Learning
Vivek Mehta, Arghya Choudhury, Utpal Roy
https://arxiv.org/abs/2508.04065 https://arxiv.org/pdf/2508…

Generalized Quantum Hadamard Test for Machine Learning
Quantum machine learning models are designed for performing learning tasks. Some quantum classifier models are proposed to assign classes of inputs based on fidelity measurements. Quantum Hadamard test is a well-known quantum algorithm for computing these fidelities. However, the basic requirement for deploying the quantum Hadamard test maps input space to L2-normalize vector space. Consequently, computed fidelities correspond to cosine similarities in mapped input space. We propose a quantum H…

CXR-CML: Improved zero-shot classification of long-tailed multi-label diseases in Chest X-Rays
Rajesh Madhipati, Sheethal Bhat, Lukas Buess, Andreas Maier
https://arxiv.org/abs/2507.19398

CXR-CML: Improved zero-shot classification of long-tailed multi-label diseases in Chest X-Rays
Chest radiography (CXR) plays a crucial role in the diagnosis of various diseases. However, the inherent class imbalance in the distribution of clinical findings presents a significant challenge for current self-supervised deep learning models. These models often fail to accurately classify long-tailed classes. Current Vision-Language models such as Contrastive Language Image Pre-training (CLIP) models effectively model the manifold distribution of the latent space, enabling high zero-shot clas…

Research on Model Parallelism and Data Parallelism Optimization Methods in Large Language Model-Based Recommendation Systems
Haowei Yang, Yu Tian, Zhongheng Yang, Zhao Wang, Chengrui Zhou, Dannier Li
https://arxiv.org/abs/2506.17551

Research on Model Parallelism and Data Parallelism Optimization Methods in Large Language Model-Based Recommendation Systems
With the rapid adoption of large language models (LLMs) in recommendation systems, the computational and communication bottlenecks caused by their massive parameter sizes and large data volumes have become increasingly prominent. This paper systematically investigates two classes of optimization methods-model parallelism and data parallelism-for distributed training of LLMs in recommendation scenarios. For model parallelism, we implement both tensor parallelism and pipeline parallelism, and int…

Modular Embedding Recomposition for Incremental Learning
Aniello Panariello, Emanuele Frascaroli, Pietro Buzzega, Lorenzo Bonicelli, Angelo Porrello, Simone Calderara
https://arxiv.org/abs/2508.16463

Modular Embedding Recomposition for Incremental Learning
The advent of pre-trained Vision-Language Models (VLMs) has significantly transformed Continual Learning (CL), mainly due to their zero-shot classification abilities. Such proficiency makes VLMs well-suited for real-world applications, enabling robust performance on novel unseen classes without requiring adaptation. However, fine-tuning remains essential when downstream tasks deviate significantly from the pre-training domain. Prior CL approaches primarily focus on preserving the zero-shot capa…

Modified non-local damage model: resolving spurious damage evolution
Roshan Philip Saji, Panos Pantidis, Mostafa E. Mobasher
https://arxiv.org/abs/2506.24099

Modified non-local damage model: resolving spurious damage evolution
Accurate prediction of damage and fracture evolution is critical for the safety design and preventive maintenance of engineering structures, however existing computational methods face significant limitations. On one hand, discrete damage and phase-field models are often computationally prohibitive for real world applications and they are less generalizable across different material classes. On the other hand, conventional gradient damage models which are based on phenomenological laws, though …

Quintessential dark energy crossing the phantom divide
Ruiqi Chen, James M. Cline, Varun Muralidharan, Benjamin Salewicz
https://arxiv.org/abs/2508.19101 https://

Quintessential dark energy crossing the phantom divide
Motivated by recent results from the DESI collaboration, we explore two classes of quintessence models that can give rise to crossing of the dark energy equation of state through the ``phantom divide'' $w=-1$. These are models with Lagrangians that involve higher powers of the kinetic energy $\dotϕ^2$, or where the dark matter (DM) mass is a function of $ϕ$. Both have similar features with respect to the reconstructed redshift-dependent $w(z)$: moderate tuning of parameters is required to ach…

Pricing Options on Forwards in Function-Valued Affine Stochastic Volatility Models
Jian He, Sven Karbach, Asma Khedher
https://arxiv.org/abs/2508.14813 https://

Pricing Options on Forwards in Function-Valued Affine Stochastic Volatility Models
We study the pricing of European-style options written on forward contracts within function-valued infinite-dimensional affine stochastic volatility models. The dynamics of the underlying forward price curves are modeled within the Heath-Jarrow-Morton-Musiela framework as solution to a stochastic partial differential equation modulated by a stochastic volatility process. We analyze two classes of affine stochastic volatility models: (i) a Gaussian model governed by a finite-rank Wishart process…

Automated Feedback on Student-Generated UML and ER Diagrams Using Large Language Models
Sebastian G\"urtl, Gloria Schimetta, David Kerschbaumer, Michael Liut, Alexander Steinmaurer
https://arxiv.org/abs/2507.23470

Automated Feedback on Student-Generated UML and ER Diagrams Using Large Language Models
UML and ER diagrams are foundational in computer science education but come with challenges for learners due to the need for abstract thinking, contextual understanding, and mastery of both syntax and semantics. These complexities are difficult to address through traditional teaching methods, which often struggle to provide scalable, personalized feedback, especially in large classes. We introduce DUET (Diagrammatic UML & ER Tutor), a prototype of an LLM-based tool, which converts a reference d…

Stable models in lower dimensions within the quadratic form of the non-metricity theory
G. G. L. Nashed, Salvatore Capozziello
https://arxiv.org/abs/2508.16679 https://

Stable models in lower dimensions within the quadratic form of the non-metricity theory
This work investigates exact charged and uncharged black hole solutions in a (2+1)-dimensional spacetime within the framework of the quadratic form of the symmetric teleparallel theory, known as $f(\mathbb{Q})$ gravity, where $\mathbb{Q}$ is the non-metricity scalar. By employing a spherically symmetric ansatz and considering both vanishing and non-vanishing electromagnetic fields, we derive new classes of black hole solutions and analyze their geometric and physical properties. The study demon…

Unsupervised Incremental Learning Using Confidence-Based Pseudo-Labels
Lucas Rakotoarivony
https://arxiv.org/abs/2508.21424 https://arxiv.org/pdf/2508.2142…

Unsupervised Incremental Learning Using Confidence-Based Pseudo-Labels
Deep learning models have achieved state-of-the-art performance in many computer vision tasks. However, in real-world scenarios, novel classes that were unseen during training often emerge, requiring models to acquire new knowledge incrementally. Class-Incremental Learning (CIL) methods enable a model to learn novel classes while retaining knowledge of previous classes. However, these methods make the strong assumption that the incremental dataset is fully labeled, which is unrealistic in pract…

Saddle Hierarchy in Dense Associative Memory
Robin Th\'eriault, Daniele Tantari
https://arxiv.org/abs/2508.19151 https://arxiv.org/pdf/2508.19151

Saddle Hierarchy in Dense Associative Memory
Dense associative memory (DAM) models have been attracting renewed attention since they were shown to be robust to adversarial examples and closely related to state-of-the-art machine learning paradigms, such as the attention mechanisms in transformers and generative diffusion models. We study a DAM built upon a three-layer Boltzmann machine with Potts hidden units, which represent data clusters and classes. Through a statistical mechanics analysis, we derive saddle-point equations that charact…

Detect Any Sound: Open-Vocabulary Sound Event Detection with Multi-Modal Queries
Pengfei Cai, Yan Song, Qing Gu, Nan Jiang, Haoyu Song, Ian McLoughlin
https://arxiv.org/abs/2507.16343

Detect Any Sound: Open-Vocabulary Sound Event Detection with Multi-Modal Queries
Most existing sound event detection~(SED) algorithms operate under a closed-set assumption, restricting their detection capabilities to predefined classes. While recent efforts have explored language-driven zero-shot SED by exploiting audio-language models, their performance is still far from satisfactory due to the lack of fine-grained alignment and cross-modal feature fusion. In this work, we propose the Detect Any Sound Model (DASM), a query-based framework for open-vocabulary SED guided by …

Modular Embedding Recomposition for Incremental Learning
Aniello Panariello, Emanuele Frascaroli, Pietro Buzzega, Lorenzo Bonicelli, Angelo Porrello, Simone Calderara
https://arxiv.org/abs/2508.16463

Modular Embedding Recomposition for Incremental Learning
The advent of pre-trained Vision-Language Models (VLMs) has significantly transformed Continual Learning (CL), mainly due to their zero-shot classification abilities. Such proficiency makes VLMs well-suited for real-world applications, enabling robust performance on novel unseen classes without requiring adaptation. However, fine-tuning remains essential when downstream tasks deviate significantly from the pre-training domain. Prior CL approaches primarily focus on preserving the zero-shot capa…

Multifractality in high-dimensional graphs induced by correlated radial disorder
David E. Logan, Sthitadhi Roy
https://arxiv.org/abs/2508.15551 https://arx…

Multifractality in high-dimensional graphs induced by correlated radial disorder
We introduce a class of models containing robust and analytically demonstrable multifractality induced by disorder correlations. Specifically, we investigate the statistics of eigenstates of disordered tight-binding models on two classes of rooted, high-dimensional graphs -- trees and hypercubes -- with a form of strong disorder correlations we term `radial disorder'. In this model, site energies on all sites equidistant from a chosen root are identical, while those at different distances are i…

ViRN: Variational Inference and Distribution Trilateration for Long-Tailed Continual Representation Learning
Hao Dai, Chong Tang, Jagmohan Chauhan
https://arxiv.org/abs/2507.17368

ViRN: Variational Inference and Distribution Trilateration for Long-Tailed Continual Representation Learning
Continual learning (CL) with long-tailed data distributions remains a critical challenge for real-world AI systems, where models must sequentially adapt to new classes while retaining knowledge of old ones, despite severe class imbalance. Existing methods struggle to balance stability and plasticity, often collapsing under extreme sample scarcity. To address this, we propose ViRN, a novel CL framework that integrates variational inference (VI) with distributional trilateration for robust long-t…

Foundation Models as Class-Incremental Learners for Dermatological Image Classification
Mohamed Elkhayat, Mohamed Mahmoud, Jamil Fayyad, Nourhan Bayasi
https://arxiv.org/abs/2507.14050

Foundation Models as Class-Incremental Learners for Dermatological Image Classification
Class-Incremental Learning (CIL) aims to learn new classes over time without forgetting previously acquired knowledge. The emergence of foundation models (FM) pretrained on large datasets presents new opportunities for CIL by offering rich, transferable representations. However, their potential for enabling incremental learning in dermatology remains largely unexplored. In this paper, we systematically evaluate frozen FMs pretrained on large-scale skin lesion datasets for CIL in dermatological …

Anisotropic flow predictions for identified and strange hadrons in $O O$ collisions at $\sqrt{s_{\mathrm{NN}}}$ = 7 TeV using model approaches
Jagbir Singh, M. U. Ashraf, A. M. Khan, S. Kabana
https://arxiv.org/abs/2507.16273

Anisotropic flow predictions for identified and strange hadrons in $O+O$ collisions at $\sqrt{s_{\mathrm{NN}}}$ = 7 TeV using model approaches
In this study, we report the predictions for the flow observables for different centrality classes in $O+O$ collisions. Our predictions utilize two different approaches, hydrodynamic and transport models, to analyze the behavior of the flow coefficients for identified ($π^\pm$, $K^\pm$ and $p (\overline{p})$) and strange ($\mathrm{K}^{0}_{\mathrm S}$, $Λ$ ($\overlineΛ$), $Ω^{-}$ ($\overlineΩ^{+}$), $Ξ^{-}$ ($\overlineΞ^{+}$), $ϕ$) hadrons. We explore particle-by-particle flow and compar…

ACT Implications for Hilltop Inflation
Monika Lynker, Rolf Schimmrigk
https://arxiv.org/abs/2507.15076 https://arxiv.org/pdf/2507.150…

ACT Implications for Hilltop Inflation
We analyze in a systematic way the implications for the classes of hilltop and hilltop-squared inflation of the recent DR6 observations by the Atacama Cosmology Telescope collaboration. We find that the reported shift in the spectral index leads to parameter ranges for these models that are significantly reduced when compared to the results obtained from the {\sc Planck} observations. We mainly focus on the more dramatic implications for the hilltop-squared class, but along the way we also high…

Empirical Evidences for the Effects of Feature Diversity in Open Set Recognition and Continual Learning
Jiawen Xu, Odej Kao
https://arxiv.org/abs/2508.13005 https://

Empirical Evidences for the Effects of Feature Diversity in Open Set Recognition and Continual Learning
Open set recognition (OSR) and continual learning are two critical challenges in machine learning, focusing respectively on detecting novel classes at inference time and updating models to incorporate the new classes. While many recent approaches have addressed these problems, particularly OSR, by heuristically promoting feature diversity, few studies have directly examined the role that feature diversity plays in tackling them. In this work, we provide empirical evidence that enhancing feature…

Interacting Scalar Field Cosmology from Full Quantum Gravity
Tom R. Ladst\"atter, Luca Marchetti
https://arxiv.org/abs/2508.16194 https://arxiv.org/pd…

Interacting Scalar Field Cosmology from Full Quantum Gravity
We study the relational cosmological dynamics emerging from interacting group field theory (GFT) models minimally coupled to a massless clock scalar field and a self-interacting scalar field. We focus on two broad classes of GFT interactions - pseudosimplicial and pseudotensorial - which generalize simplicial and tensorial interactions, respectively. Treating these interactions perturbatively, we extract the effective cosmological dynamics using mean-field techniques. In the geometric sector, w…

Multiscale Video Transformers for Class Agnostic Segmentation in Autonomous Driving
Leila Cheshmi, Mennatullah Siam
https://arxiv.org/abs/2508.14729 https://

Multiscale Video Transformers for Class Agnostic Segmentation in Autonomous Driving
Ensuring safety in autonomous driving is a complex challenge requiring handling unknown objects and unforeseen driving scenarios. We develop multiscale video transformers capable of detecting unknown objects using only motion cues. Video semantic and panoptic segmentation often relies on known classes seen during training, overlooking novel categories. Recent visual grounding with large language models is computationally expensive, especially for pixel-level output. We propose an efficient vide…

RICO: Two Realistic Benchmarks and an In-Depth Analysis for Incremental Learning in Object Detection
Matthias Neuwirth-Trapp, Maarten Bieshaar, Danda Pani Paudel, Luc Van Gool
https://arxiv.org/abs/2508.13878

RICO: Two Realistic Benchmarks and an In-Depth Analysis for Incremental Learning in Object Detection
Incremental Learning (IL) trains models sequentially on new data without full retraining, offering privacy, efficiency, and scalability. IL must balance adaptability to new data with retention of old knowledge. However, evaluations often rely on synthetic, simplified benchmarks, obscuring real-world IL performance. To address this, we introduce two Realistic Incremental Object Detection Benchmarks (RICO): Domain RICO (D-RICO) features domain shifts with a fixed class set, and Expanding-Classes …