iiinova — where your digital self keeps evolving

Erkan Karabulut @erkankarabulut.bsky.social · 23/08/2026

Time is a resource that you can spend on your to-do list to have big numbers in life. A balanced to-do list contains many leisure (maintenance) times scheduled tightly with everything else. The total (big) number of events, leisure or not, is to be optimized. Religiously healthy human. Sarcasm.

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 03/07/2026

v1.1 PyAerial provides a neural approximation to co-occurrence counting to scale standard rule miners and address rule explosion. Any rule miner can be implemented on top of PyAerial architecture. It now supports FP-Growth-like rule extraction and can learn rules of arbitrary length much faster!

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 03/07/2026

PyAerial can now do much faster knowledge discovery from tabular data via rule learning with a tree-based rule extractor (v1.1)! If you are interested in knowledge discovery or interpretable machine learning for high-stakes decision making, check out PyAerial. github.com/DiTEC-projec...

github.com

GitHub - DiTEC-project/pyaerial: Scalable association rule mining from tabular datasets.

Scalable association rule mining from tabular datasets. - DiTEC-project/pyaerial

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 30/04/2026

#IJCAI2026, #ECAI2026 decisions are out. Despite having written a rebuttal in hard personal circumstances, receiving no response to our rebuttal from reviewers, SPCs, ACs, or PCs is really heartbreaking.

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

📊 Results (3/3) TFM-learned rules are more generalizable to unseen data in comparison to algorithmic or neurosymbolic methods. Co-authored with @dfdazac.bsky.social, @p-groth.bsky.social, Martijn C Schut, and @vdegeler.bsky.social. 🧵7/7

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

📊 Results (2/3) TFMs can scale up to mid-size tables (~50 columns, ~10.000 rows) under a minute on affordable resources for association rule learning. However, on larger datasets, neurosymbolic methods (PyAerial) are still the only scalable approach. 🧵6/7

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

📊 Results (1/3) TFMs with our TabProbe can address both rule explosion and remain robust in low data scenarios in terms of statistical rule quality as well as in interpretable rule-based classification tasks. 🧵5/7

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

📖 The research gap (2/2) Neurosymbolic methods, such as our PyAerial, can address rule explosion and scale to larger datasets. However, neurosymbolic methods suffer from degraded performance in low-data scenarios. 🧵5/7

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

📖 The research gap (1/2) Algorithmic rule miners, without an effective search space reduction (which requires heuristics over the data), lead to rule explosion, which is an excessive number of rules that are hard to interpret and post-process. 🧵4/7

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

❓What's new? (2/2) TabProbe - We instantiated our framework with TFMs with a classification objective (TabPFNv2.5, TabICL, and TabDPT), and showed that TFMs can be re-purposed to learn association rules out-of-the-box with no additional training and without frequent itemset mining! 🧵3/7

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

📜 Preprint: arxiv.org/pdf/2602.146... 🐍 Code: github.com/DiTEC-projec... ❓What's new? (1/2) A model-agnostic rule learning framework - We show that any conditional probabilistic model that supports querying for features conditioned on others can be re-purposed to learn association rules. 🧵2/7

arxiv.org

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 19/02/2026

Tabular foundation models can learn association rules out of the box! For the non-expert: If you work with tables and want to discover interesting patterns, tabular foundation models (TFMs) can quickly find significant, interesting, and generalizable patterns with no additional training. 🧵1/7

101 0

Reposted by Erkan Karabulut

BJGP @bjgp.org · 26/01/2026

No. 2 #BJGPTop10 AI in lung cancer detection: AI algorithm identifies risk of lung cancer 4 months earlier doi.org/10.3399/BJGP... #LungCancer #CancerResearch #AIHealthcare #DigitalHealth #EarlyDetection #MedicalInnovation #ArtificialIntelligence #PrimaryCare #BJGP #research

012 0

Reposted by Erkan Karabulut

Madelon Hulsebos @madelonhulsebos.bsky.social · 22/01/2026

Join the TRL Seminar tomorrow for talks on query ambiguity in open-domain data analysis @daniel-gomm.bsky.social, the SQaLE dataset for training text-to-SQL models @cowolff.bsky.social, and pattern/rule inference from tabular data @erkankarabulut.bsky.social. Info: trl-lab.github.io/trl-seminar/

trl-lab.github.io

TRL Seminar | TRL Lab

153 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/11/2025

PyAerial documentation is now much more comprehensive and has a lot of examples: pyaerial.readthedocs.io/en/latest/in... If you are interested in interpretable machine learning or knowledge discovery, check out PyAerial!

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 12/11/2025

Yesterday, I gave a seminar on Scalable Knowledge Discovery with Neurosymbolic Rule Learning at the University of Amsterdam’s Data Science Center. Here’s a blog post with code samples and experiments on knowledge discovery from tabular data: 📄 erkankarabulut.github.io/blog/uva-dsc....

erkankarabulut.github.io

UvA Data Science Center Seminar: Scalable Knowledge Discovery

A walkthrough of scalable knowledge discovery from tabular data using neurosymbolic methods, covering association rule mining, its challenges, and neurosymbolic solutions with Aerial.

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/10/2025

Another question came to my mind about this definition: can such consciousness be transferred to a new body? Or if one’s body were cryopreserved, would it retain the same consciousness—would it still be the same person? Practically, we’d want consciousness to be transferable.

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/10/2025

The more you are aware (I guess of yourself, others, and the connections), the higher your consciousness level is. So a baby (or an AI), over time, learns more about itself, others, and its surroundings, and by doing so gets more conscious.

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/10/2025

One particularly interesting view I learned was from Psychologist Markus Ofner. Summarizing with my own words: Consciousness is in the air and in us. We don't own it, but we get conscious. We are all individuals, but also part of a whole, all interconnected.

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/10/2025

Similar to the question of "do numbers (or math) exist in nature or did we invent it?", does consciousness exist in nature or should we invent it? Assuming that we can't define it, hence understand what it is, can't we come up with a definition that works better for us, similar to math?

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/10/2025

Why is defining consciousness not trivial? I am attending the ECAI2025 workshop ACAI - Awakening Consciousness in Artificial Intelligence. And hearing different sound and consistent, at first, definitions or assumptions around consciousness made me think of this question.

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 22/10/2025

If you're attending and interested in knowledge discovery, interpretable machine learning, or XAI, say hi! 📜 arxiv.org/pdf/2509.20113 🐍 tinyurl.com/3z8cmuhw 🐍 Python Library: github.com/DiTEC-projec... Co-authored with @dfdazac.bsky.social, @p-groth.bsky.social, and @vdegeler.bsky.social. 🧵2/2

arxiv.org

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 22/10/2025

I'm excited to present our new method for enhancing knowledge discovery from tabular data using tabular foundation models at the #ECAI2025 conference workshops (ANSyA) this Sunday, October 26th. See our paper, Discovering Association Rules in High-Dimensional Small Tabular Data, below 👇. 🧵1/2

Discovering Association Rules in High-Dimensional Small Tabular Data

111 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

🧠 Neurosymbolic rule learning methods start a paradigm shift in knowledge discovery by allowing us to utilize prior knowledge while discovering new knowledge! Together with @dfdazac.bsky.social, @p-groth.bsky.social, and @vdegeler.bsky.social. 🧵8/8

010 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

📊 On 5 real-world gene expression datasets with 18K+ columns and <100 rows, both methods led to significantly higher quality association rules in terms of confidence and association strength, with limited increase in execution time (1.2 - 2 times at most). 🧵7/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

1️⃣ Aerial+WI. Weight initialization of Aerial+ based on tabular data embeddings from a foundation model, using a projection encoder. 2️⃣ Aerial+DL. Tabular embeddings are aligned with Aerial+'s reconstructions via a projection encoder and joint loss, ensuring a better semantic column alignment. 🧵6/8

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

💡In the scope of knowledge discovery via rule learning, we adapted and evaluated two transfer learning methods that utilize table embeddings from a tabular foundation model, TabPFN. 🧵5/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

🔍 Therefore, we introduce the problem of discovering association rules in high-dimensional small tabular data for the first time. 📗Tabular foundation models have addressed this issue by pre-training on large datasets and transfer learning to small datasets for predictive tasks. 🧵4/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

❓However, Neurosymbolic methods inherit limitations of neural networks into rule learning, particularly reduced performance in low-data regimes. In rule learning, this translates to not being able to capture high-quality patterns in the data. 🧵3/8

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

🐍 Library: github.com/DiTEC-projec... 📊 We show that knowledge discovery from high-dimensional tables, as in gene expression datasets (~18K columns), is scalable with Neurosymbolic rule learning, Aerial+, a method we have proposed earlier (arxiv.org/pdf/2504.19354, presented at NeSy 2025). 🧵2/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 26/09/2025

Tabular foundation models can boost knowledge discovery from high-dimensional tabular data with few instances (e.g., gene expression or rare disease data, 1K+ columns and <100 rows)! 📄 Short Paper (Accepted at ECAI 2025 workshops, ANSyA): arxiv.org/pdf/2509.20113 🐍 Code: tinyurl.com/3z8cmuhw 🧵1/8

lnkd.in

This link will take you to a page that’s not on LinkedIn

131 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

PyAerial runs 1–2 orders of magnitude faster than major rule mining libraries in C/C++, R, and Python. Using an under-complete Autoencoder, it avoids non-informative patterns (rule explosion), captures most significant associations, higher confidence, stronger links, and full coverage. 🧵8/8

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

It can also be integrated with rule visualization software(s), such as NiaARM. (coming out soon) PyAerial supports transfer learning from a tabular foundation model (such as TabPFN), e.g., by re-using model weights, or semantical alignment to a given set of table embeddings. 🧵7/8

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

Rules of different forms, e.g., classification rules, item constraints, can be learned as part of the exposed Aerial+ interfaces. PyAerial can run on parallel threads and on a GPU, and scales on high-dimensional tables (1K+ columns). 🧵6/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

🎯 What can it do (technical)? In just 2 lines of Python code, PyAerial can learn a concise set of high-quality association rules from a table in pandas dataframe form, by utilizing an under-complete denoising autoencoder! 🧵5/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

For high-stakes decisions (e.g., recidivism), PyAerial learns associations between features and labels as classification rules. Rather than scanning the entire dataset, PyAerial lets users specify items of interest, improving runtime and the interoperability of discovered patterns. 🧵4/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

On high-dimensional tables (1K+ columns) like gene expression datasets, PyAerial runs 1–2 orders of magnitude faster than leading rule mining libraries. In low-data regimes (e.g., rare diseases), PyAerial can leverage tabular foundation models (TabPFN) to boost performance (paper out soon!). 🧵3/8

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

🔧 Practical implications PyAerial can extract high-quality, significant patterns from any given table, without causing the well-known rule explosion problem (many redundant, non-informative patterns). 🧵2/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 14/09/2025

New software paper published in the SoftwareX journal, describing our neurosymbolic scalable association rule learner PyAerial for knowledge discovery and interpretable inference! 🔎 See the practical implications, capabilities, and benchmarking below! 📜 lnkd.in/eQubA7MD 🐍 lnkd.in/eDWnVWFr 🧵1/8

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 09/09/2025

Tomorrow at the Neurosymbolic Learning and Reasoning (NeSy2025) conference, I will be presenting our novel, scalable association rule 'learning' method for tabular data. Say hi if you are interested in knowledge discovery and/or interpretable ML! 📜 tinyurl.com/48fmu3eh 🐍 tinyurl.com/45s75r6w

010 0

Erkan Karabulut @erkankarabulut.bsky.social · 07/09/2025

I will be at the NeSy2025 conference this week, presenting Aerial scalable neurosymbolic tabular rule learner! Feel free to reach out if you are interested in knowledge discovery, interpretable inference!

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 15/08/2025

PyAerial looks much nicer now, with new CI/CD implemented, 50+ test cases and visualizations of association rules! Learning rules over large tables for knowledge discovery and interpretable inference is much easier with PyAerial. github.com/DiTEC-projec...

github.com

GitHub - DiTEC-project/pyaerial: A Python package of the Aerial neurosymbolic association rule mining algorithm from tabular datasets.

A Python package of the Aerial neurosymbolic association rule mining algorithm from tabular datasets. - DiTEC-project/pyaerial

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 30/07/2025

I wrote a new blog post that introduce our new neurosymbolic tabular rule learner (accepted at NeSy 2025) in an intuitive level. Feel free to reach out if you wanna discuss about it! 📄 erkankarabulut.github.io/blog/scalabl...

erkankarabulut.github.io

030 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

🐍 Open-source We provide a super-easy-to-use Python library for Aerial+ on categorical tabular data. In 2 lines of code, you can train a model and extract rules. Every aspect of our model and algorithm is parameterized. We welcome every bit of contribution! 🐍 github.com/DiTEC-projec... 🧵8/8

github.com

GitHub - DiTEC-project/pyaerial: A Python package of the Aerial neurosymbolic association rule mining algorithm from tabular datasets.

A Python package of the Aerial neurosymbolic association rule mining algorithm from tabular datasets. - DiTEC-project/pyaerial

000 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

📊 Key Insights ➡️ The proposed method is orthogonal to many existing solutions to the rule explosion problem. ➡️ Aerial+ can potentially be incorporated into larger neural networks for better interpretability. 🧵7/8

110 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

📊 Key Insights ➡️ Neurosymbolic rule learning can tackle rule explosion. It leads to a concise set of high-quality association rules with full data coverage. ➡️ Learned rules significantly boosted downstream interpretable inference tasks in execution time with equal or higher accuracy. 🧵6/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

✅ Aerial+ Rule extraction If a forward run on the trained model with a set of marked categories A results in successful reconstruction (high probability) of categories C, we say that marked features A imply the successfully reconstructed features C, such that A → C \ A (no self-implication). 🧵5/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

✅ Method: Aerial+ Learn a compact neural representation of the tabular data using an under-complete denoising Autoencoder. Extract association rules from the trained Autoencoder by exploiting its reconstruction mechanism 🧵4/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

🔎 Challenge Existing rule miners from tabular data are mostly algorithmic or optimization-based, which results in rule explosion and impractical execution times without effective search space reduction. Search space reduction --> item constraints, top-k rules, and many more strategies. 🧵3/8

100 0

Erkan Karabulut @erkankarabulut.bsky.social · 23/07/2025

📗Background Rule learning discovers patterns from data and represents them as formal logical statements. Patterns referring to, e.g., frequent co-occurrences or non-occurrences (negations), high-utility data points, strong associations, OR statements, etc. 🧵2/8

100 0