We learn more expressive mixture models that can subtract probability density by squaring them. 🚨We show squaring can reduce expressiveness. To tackle this we build sum of squares circuits🆘. 🚀We explain why complex parameters help, and show an expressiveness hierarchy around🆘

RT @ema_marconato: 🧵Why are linear properties so ubiquitous in LLM representations?. We explore this question through the lens of 𝗶𝗱𝗲𝗻𝘁𝗶𝗳𝗶𝗮….

RT @EmilevanKrieken: We propose Neurosymbolic Diffusion Models! We find diffusion is especially compelling for neurosymbolic approaches, co….

RT @LennertDS_: Just under 10 days left to submit your latest endeavours in ⚡#tractable⚡ probabilistic models❗. Join us at TPM @auai.org #U….

RT @diegocalanzone: In LoCo-LMs, we propose a neuro-symbolic loss function to fine-tune a LM to acquire logically consistent knowledge fro….

RT @jjcmoon: We developed a library to make logical reasoning embarrassingly parallel on the GPU. For those at ICLR 🇸🇬: you can get the….

RT @e_giunchiglia: 🚨New at #ICLR: we introduce the first ever 𝐥𝐚𝐲𝐞𝐫 that makes 𝐚𝐧𝐲 neural network 𝐜𝐨𝐦𝐩𝐥𝐢𝐚𝐧𝐭 𝐛𝐲 𝐝𝐞𝐬𝐢𝐠𝐧 with constraints expr….

RT @pascalhitzler:

RT @PMinervini: @chaitjo @benfinkelshtein @ffabffrasca @mmbronstein @phanein @michael_galkin @chrsmrrs hi, we found problematic benchmarks….

RT @bkailkhu: Our paper "Low-rank finetuning for LLMs is inherently unfair" won a 𝐛𝐞𝐬𝐭 𝐩𝐚𝐩𝐞𝐫 𝐚𝐰𝐚𝐫𝐝 at the @RealAAAI colorai workshop! #AAAI….

You can find the speakers bios and the abstracts of presentations here: Check them out!.

The last speaker of the workshop is Alexandros Georgiou, who is giving an introduction to polynomial networks and equivariant tensor network architecture, as well as how to implement them.

After lunch break, Andrew G. Wilson (@andrewgwils) is now giving his presentation on the importance of linear algebra structures in ML, as well as on how to navigate such structures in practice.

After Nadav it is now the turn of Guillaume Rabusseau @grwip, who is joining us online. Guillaume guides us through interesting expressiveness relationships of families of RNNs that are parameterized through tensor factorizations techniques

Live from the CoLoRAI workshop at AAAI. @nadavcohen is now giving his talk on "What Makes Data Suitable for Deep Learning?".Tools from quantum physics are shown to be useful in building more expressive deep learning models by changing the data distribution

RT @jjcmoon: We all know backpropagation can calculate gradients, but it can do much more than that!. Come to my #AAAI2025 oral tomorrow (1….

RT @benjiewang_cs: Circuits use sum-product computation graphs to model probability densities. But how do we ensure the non-negativity of t….

We are going to present our poster "Sum of Squares Circuits" at AAAI in Philadelphia today. Hall E 12:30pm-14:00pm poster #840. We trace expressiveness connections of different types of additive and subtractive deep mixture models and tensor networks. 📜

RT @GabVenturato: 🔥 Can AI reason over time while following logical rules in relational domains? We will present Relational Neurosymbolic M….

RT @piermonn: These last three weeks, I had the chance to be back at Università degli Studi di Bari as an invited researcher, to continue w….

RT @kerstingAIML: 🚀 Meet NeST, the first neuro-symbolic transpiler! It converts SPLL, a novel probabilistic language, into code across #AI….