DEV Community: Centaur AI

Neuro-Symbolic AI: From Research to Industry

Can Mingir — Wed, 10 Dec 2025 17:11:52 +0000

📘 This week:
Neuro-Symbolic AI brings together the statistical nature of machine learning with the formal reasoning capabilities of symbolic AI. It seeks to offer a balanced approach to contemporary AI technologies, by combining the ability to learn from data, with the capacity to reason upon knowledge acquired from an environment. The main criticism of neural machine learning lies in its lack of explainability and semantics, which are key requirements in safety-critical applications, yet inherent strengths of logic-based methods. Recently, several corporations have publicly announced products and technologies grounded in Neuro-Symbolic AI methodologies. This talk provided a concise review of the foundations, frameworks and tools underlying Neuro-Symbolic AI, along with illustrative applications. It concludes by highlighting current trends and research directions in the field.

📝 Suggested paper:
A Garcez, LC Lamb:
Neurosymbolic AI: the 3rd wave. Artif. Intell. Review 56(11):12387-12406 (2023)

https://link.springer.com/article/10.1007/s10462-023-10448-w

Training Large Reasoning Models

Can Mingir — Wed, 03 Dec 2025 16:22:49 +0000

📘 This week, Dr. Asim Munawar from IBM’s Watson Research will examine how integrating Neuro-Symbolic AI approaches with large language models (LLMs) can enhance their reasoning, interpretability, and reliability.

Dr. Asim Munawar is a Project Lead at IBM’s Watson Research Center in New York, where he heads efforts to enhance reasoning, planning, and agentic workflows in enterprise-scale large language models. With over 15 years of experience in AI – more than a decade of it at IBM Research – he has held key leadership roles, including Manager and Program Director for Neuro-Symbolic AI https://asimmunawar.github.io

This presentation will explore architectures that combine symbolic logic, structured representations, and deep learning, and outlines strategies to address limitations such as brittleness, hallucination, and lack of transparency in current LLMs.

📝 Bonus: The DeepSeek paper will be covered in Asim's presentation.
https://arxiv.org/abs/2501.12948

Can/Will LLMs Learn to Reason?

Can Mingir — Wed, 12 Nov 2025 16:34:14 +0000

Join us for an interactive session exploring Neuro-Symbolic AI. This week, Abulhair Saparov from Purdue University is presenting “Can/Will LLMs Learn to Reason?”

This week:

Reasoning—the process of drawing conclusions from prior knowledge—is a hallmark of intelligence. Large language models, and more recently, large reasoning models have demonstrated impressive results on many reasoning-intensive benchmarks. Careful studies over the past few years have revealed that LLMs may exhibit some reasoning behavior, and larger models tend to do better on reasoning tasks. However, even the largest current models still struggle on various kinds of reasoning problems. In this talk, we will try to address the question: Are the observed reasoning limitations of LLMs fundamental in nature? Or will they be resolved by further increasing the size and data of these models, or by better techniques for training them? I will describe recent work to tackle this question from several different angles.

📢 New Series: Neuro-Symbolic Wednesday

Can Mingir — Wed, 05 Nov 2025 08:22:27 +0000

Join us for an interactive session exploring Neuro-Symbolic AI, the emerging paradigm that blends the strengths of neural networks with symbolic reasoning. We will discuss how hybrid approaches can enhance generalization, interpretability, and reasoning, and how these methods are shaping the future of intelligent systems.

Whether you’re a researcher, engineer, or simply curious about the cutting edge of AI, you’ll find an engaging space to learn, connect, and exchange ideas.

Event Format:
🔍 Reading Group
🗣️ Guest Spotlights
💬 Research Roundtable
🛠️ Open-Source Review
🤝 Collaboration Hour

RSVP at Luma

📢 Neuro-Symbolic AI Summer School 2025 | Online Event | Aug 14 - 15

Can Mingir — Wed, 13 Aug 2025 18:43:42 +0000

http://lu.ma/pqzv80yd?utm_source=devto

📅 Agenda

Day 1: Frameworks and Foundations | AUG 14

📜 Big Picture
9:00–9:30 AM ET – Alexander Gray, Centaur AI Institute – Neuro-Symbolic AI and NSSS4 (Overview)
9:30–9:40 AM – Q&A / Break
9:40–10:20 AM – Arvind Narayanan, Princeton – Assessing the Current State of AI (Talk)
10:20–10:30 AM – Q&A / Break

🏗 Frameworks
10:30–11:05 AM – Leilani Gilpin, UC Santa Cruz – Neuro-Symbolic AI for Safer Autonomous Vehicles (Tutorial)
11:05–11:15 AM – Q&A / Break
11:15–12:00 PM – Mateo Zarlenga & Pietro Barbiero, Cambridge – Foundations of Interpretable Models (Tutorial)
12:00–12:10 PM – Q&A / Break
12:10–12:55 PM – Parisa Kordjamshidi, Michigan State – Compositional Learning in Language and Vision (Tutorial)
12:55–1:05 PM – Q&A / Break

📐 Mathematical Foundations
1:05–1:45 PM – Parikshit Ram, IBM Research – How to Measure Compositionality, and Why it Leads to Better Generalization (Talk)
1:45–1:55 PM – Q&A / Break
1:55–2:35 PM – Peihao Wang, UT Austin – Why Neural Networks Can Discover Symbolic Structures (Talk)
2:35–2:45 PM – Q&A / Break
2:45–3:25 PM – Changlong Wu, Purdue – Why Current Models Will Always Hallucinate (and a Path Forward) (Talk)
3:25–3:35 PM – Q&A / Break
3:35–4:35 PM – Sridhar Mahadevan, Adobe Research & UMass Amherst – Category Theory: The Mathematics of Symbolic Structures (Tutorial)
4:35–4:45 PM – Q&A / Break

Day 2: Methods and Systems | AUG 15

🤖 Neuro-Symbolic AI Software
9:00–9:40 AM – Bowen Li, Carnegie Mellon – Generalizing to New Situations in Robotics (Talk)
9:40–9:50 AM – Q&A / Break
9:50–10:30 AM – Olga Vileskaia & Kevin O'Connor, Centaur AI Institute – Explainability While Retaining Predictive Accuracy (Talk)
10:30–10:40 AM – Q&A / Break

🧩 Learning Symbolic Models
10:40–11:20 AM – Bin Yu, UC Berkeley – Interpretable and Veridical Data Science (Talk)
11:20–11:30 AM – Q&A / Break
11:30–12:10 PM – Felix Petersen, Stanford – Deep Differentiable Logic Gate Networks (Talk)
12:10–12:20 PM – Q&A / Break
12:20–1:00 PM – Hikaru Shindo, TU Darmstadt – Neuro-symbolic Agentic Systems (Talk)
1:00–1:10 PM – Q&A / Break

🛡 Safer AI Systems
1:10–1:50 PM – Soroush Saghafian, Harvard – Human + AI "Centaur" Systems (Talk)
1:50–2:00 PM – Q&A / Break
2:00–2:40 PM – Pranava Madhyastha, City Univ London – New Results in Controllable Text Generation (Talk)
2:40–2:50 PM – Q&A / Break

➗ AI Systems for Mathematics
2:50–3:30 PM – Shange Tang, Princeton – State-of-the-art Performance in Automated Mathematical Theorem Proving (Talk)
3:30–3:40 PM – Q&A / Break
3:40–4:20 PM – Ankit Anand, DeepMind – Curious Case of AI in Maths: Being Proficient in Advancing Open Conjectures in Maths Yet Having Struggles in AI for Education (Talk)
4:20–4:30 PM – Q&A / Break

🔮 Looking Forward
4:30–5:30 PM – Panel on The Future of AI – Rich Sutton, Univ Alberta; Leonardo de Moura, Amazon; Artur Garcez, City Univ London; more TBA; moderator: Alexander Gray – Discussion including open Q&A
5:30–5:35 PM – Alexander Gray, Centaur AI Institute – What's Coming Next and How to Participate (Closing remarks)