Michelangelo Puliga

🚀 New Open Access Article 🚀 Emanuele Cavenaghi published his work on TORS! 👇 His paper, "Towards a Causal Decision-Making Framework for Recommender Systems", addresses the limitations of learning from offline observed data and the challenges of current debiasing strategies. To advance the field, he proposes integrating causality frameworks such as potential outcomes and structural causal models into recommender systems. This approach, developed in collaboration with Free University of Bozen-Bolzano, aims to enhance future research by applying causal discovery strategies on offline data, enabling the computation of counterfactuals and improving debiasing methods. You can find the full paper at: https://lnkd.in/dVgcJbMq Congratulations to Emanuele and his team for this amazing contribution! Dive into our research to explore these innovative ideas and their potential to transform recommender systems at: https://lnkd.in/dtpqzmX4 #Research #ArtificialIntelligence #MachineLearning #RecommenderSystems #Causality #MADLab

9

What is even getter that an LLM with a RAG? RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval) takes Retrieval-Augmented Generation (RAG) to the next level: RAPTOR is like equipping an LLM with a multi-level map for navigating the knowledge landscape. Traditional RAG is akin to providing an LLM with a flat, single-layered map of the information terrain. While helpful, this map lacks depth and context, making it challenging for the LLM to grasp the relationships between different regions and navigate complex queries efficiently. RAPTOR, on the other hand, equips the LLM with a multi-level, interactive map that offers both high-level overviews and granular details. By recursively clustering and summarizing information into a hierarchical tree structure, RAPTOR creates a map with multiple zoom levels, allowing the LLM to seamlessly switch between broad context and specific details as needed. This empowers the LLM to navigate even the most intricate knowledge landscapes with ease, delivering accurate and contextually rich responses to user queries.

15

Indirect Comparisons and Network Meta-Analyses To learn how to implement #economic_evaluation_techniques (#CEA and #CUA) using decision analytic models, including #decision_trees and #Markov_models in a one-to-one setting: https://lnkd.in/dUcpjq6p.

53

Knolwedge graphs can help improve your RAG solutions by providing you with the ability to answer thematic questions across your dataset. I haven't yet checked out LazyGraphRAG, but GraphRAG has some very cool applications. Thanks for sharing, Aishwarya!

26

#Introducing #Positron: The Ultimate #Data Science #IDE for #R and #Python Lovers 🚀 I just tried out Positron, the brand new IDE by Posit (formerly #RStudio), and I'm in love! 😍 As someone who's been using R in VS Code, I always felt frustrated by the lack of #RStudio like functionality. But Positron has changed the game. 🎮 This next-generation #data science IDE is built on top of the open-source version of VS Code optimizing the #UI for data science workflows. 🧑‍💻 It seamlessly supports both R and Python, with no need to install pesky extensions. 🐍 Positron comes packed with features that make coding a breeze, like a built-in data and variable explorer, easy-to-use keyboard shortcuts, and the ability to run Python scripts line by line with a single click. 🔥 If you're an R or Python enthusiast looking for an IDE that combines the best of both worlds, Positron is the answer. 🙌 Give it a try and let me know what you think! #Positron #RStudio #DataScience #IDE #Python #Vscode

2

New custom dataset in the block to enhance discovery and development of carbon-capturing materials. https://lnkd.in/djBp2Cwb What's it about? This study is focused on improving Direct Air Capture, a technology that pulls carbon dioxide (CO2) from the air to combat climate change. Specifically, the research explores metal−organic frameworks (MOFs), which are materials that can trap CO2 molecules. The goal is to find MOFs that can efficiently capture CO2, even in humid conditions, which is challenging. Why MOFs? MOFs are like tiny, intricate sponges with a huge surface area packed into a small volume. They can be designed at a molecular level to capture specific molecules like CO2. This flexibility makes them potentially powerful tools for removing CO2 from the atmosphere. What did the researchers do? They used quantum chemistry calculations to predict how thousands of different MOFs react with CO2 and water (H2O). By combining these calculations with machine learning models, they could speed up the search for the best MOF designs. Their dataset includes over 38 million calculations and covers thousands of MOFs, making it one of the most comprehensive of its kind. Why is this important? With climate change accelerating, we need effective ways to reduce atmospheric CO2. DAC is one method, but it requires materials that can capture CO2 efficiently and withstand environmental factors like humidity. This research helps by narrowing down the vast number of possible MOFs to those most likely to be effective, saving time and resources in the fight against climate change. In essence, the article discusses a significant step forward in identifying new materials that could help us more efficiently remove CO2 from the atmosphere, a critical action to mitigate global warming and its impact. #climatechange #artificialintelligence #machinelearning #carbonmarket

1

1 commento

Really #cool and very #powerful💯🍀 Awesome approach to simplify #ontology <-> #sql #workflow and #pipelines

3

📈 Dataframe libraries evolving faster than you can keep up? Marco Gorelli's Data Umbrella webinar is your cheat code! Learn Polars expressions and how to build library-agnostic data tools. 👀 Watch now: https://lnkd.in/dpMGfx4a #DataScience #TechTips

7

1 commento

Geometric Learning via data-efficient PDE-G-CNNs Remco Duits https://lnkd.in/eGRtXxge Abstract: We consider PDE-based Group Convolutional Neural Networks (PDE-G-CNNs) that generalize Group equivariant Convolutional Neural Networks (G-CNNs). In PDE-G-CNNs a network layer is a set of PDE-solvers. The underlying (non)linear PDEs are defined on the homogeneous space M(d) of positions and orientations within the roto-translation group SE(d) and provide a geometric design of the roto-translation equivariant neural network. The network consists of morphological convolutions with (approximative) kernels solving nonlinear PDEs (HJB equations for max-pooling over Riemannian balls), and linear convolutions solving linear PDEs (convection, fractional diffusion). Our analytic approximation kernels are accurate in comparison to our recent exact PDE-kernels. Common mystifying (ReLU) nonlinearities are now obsolete and excluded. We achieve high data-efficiency of our networks: better classification results in image processing with both less training data and less network complexity. Moreover, we have network interpretability as we train sparse association fields (modeling contour perception in our own visual system).

7

Avoiding #Bias in Observational Studies To learn how to implement #economic_evaluation_techniques (#CEA and #CUA) using decision analytic models, including #decision_trees and #Markov_models in a one-to-one setting: https://lnkd.in/dUcpjq6p.

64

Meaning at the Planck scale? Contextualized word embeddings for doing history, philosophy, and sociology of science | by Arno Simons | ArXiv This paper explores the potential of contextualized word embeddings (CWEs) as a new tool in the history, philosophy, and sociology of science (HPSS) for studying contextual and evolving meanings of scientific concepts.  The study underscores the importance of domain-specific pretraining for analyzing scientific language and demonstrates the cost-effectiveness of adapting pretrained models for HPSS research. By offering a scalable and transferable method for modeling the meanings of scientific concepts, CWEs open up new avenues for investigating the socio-historical dynamics of scientific discourses. (...)  Contextualized word embeddings (CWEs) provide a significant leap forward. CWEs rely on patterns of usage in large text corpora to encode words as points in a continuous vector space. In this space, semantic and syntactic relationships between words are represented through relative distances, with similar contexts grouped closely together. Unlike earlier word embeddings, which assign static representations to words, CWEs dynamically adjust a word’s representation depending on its context. This adaptability enables researchers to analyze how meanings differ and evolve across time, disciplines, or other settings, surpassing the static associations revealed by traditional methods https://lnkd.in/ekv_8cfq #AI #LLMs #scientific #concepts #CWEs #history #philosophy #sociology of #science #word #sense #disambiguation #induction #lexical #semantic #change #discourse #analysis #language #meaning #computation #Planck #scale

1

If we connected recently at FOSS4G 2024 Belém, I have a #YouTube channel where I collect past #talks and #workshops on #geospatial #datascience and #geostatistical #modeling. If you are interested in these topics, consider subscribing. Below is a talk from 2021 on #geostatistical #learning, which is the formalization of #machine #learning with #geospatial #data: https://lnkd.in/dsjyXVRE

38

A class of kernel-based scalable algorithms for data science We are quite excited to share this pre-print ([*EDIT* : included numerical examples] :https://lnkd.in/eda4jTEG), [*EDIT* : included notebook files: https://lnkd.in/eQR9Psiq], discussing a series of scalable algorithms for RKHS methods (see https://lnkd.in/gmjKPPS8), suitable to large scale dataset. We are still testing this approach, however a first implementation can be found in our RKHS library codpy (https://lnkd.in/e4kUprjY), that is our main numerical tool to design applications (https://lnkd.in/e-tSzz55). Here is the abstract : We present several generative and predictive algorithms based on the RKHS (reproducing kernel Hilbert spaces) methodology, which most importantly are scalables. It is well recognized that the RKHS methodology leads one to efficient and robust algorithms for numerous tasks in data science, statistics, and scientific computations. However, the standard implementations remain difficult to scale to encompass large data sets. In this paper, we introduce a simple and robust, divide-and-conquer approach which applies to large scale data sets and relies on suitably formulated, kernel-based algorithms: we distinguish between extrapolation, interpolation, and optimal transport steps. We explain how to select the best algorithms in specific applications thanks to some feedback, mainly consisting of perfomance criteria. We mainly target applications and challenging problems arising in industrial applications, such as generation of mesh for efficient numerical simulations, design of generators of conditional distributions, transition probability matrix for statistic or stochastic applications, as well as various tasks of interest to the artificial intelligence community. Indeed, the proposed algorithms are relevant for supervised or unsupervised learning, generative methods, and reinforcement learning. #RKHS #AI #LargeScaleDataset #ML #clustering #Sinkhorn #GromovWasserstein

40

4 commenti

📣 Don't miss Marco Gorelli's talk at #PyConItaly on supporting multiple DataFrame libraries seamlessly.🙌 He'll cover the "just convert to pandas" limitations and how the DataFrame Interchange Protocol and Narwhals offer a better solution for robustness and efficiency. 🚀 https://2024.pycon.it/en #DataFrame #Polars #OpenSource

1

What will science be like when we use ever more machine learning? Will it just be all automated without humans? Data in, scientific results out? On the contrary. We need scientists to interpret models, add (causal) domain knowledge, and understand sources of uncertainty. That's why Timo and I wrote Supervised Machine Learning for Science. You can read the book here for free: https://lnkd.in/gumhVS-X

463

20 commenti

🚀 Introducing 𝗔𝘂𝘁𝗼𝗚𝗚𝗨𝗙: An Automated Graphical Interface for GGUF Model Quantization 🎉 AutoGGUF, a new graphical user interface (PyQt6) app developed in Python to streamline the quantization process of GGUF models using the llama.cpp library. 🎳 𝗞𝗲𝘆 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀: - Automated Download & Management: Seamlessly manage llama.cpp backends, including CUDA. - Simple Model Selection & Quantization: Easily choose and quantize models. - Configurable Parameters: Tailor your quantization settings. - Resource Monitoring: Monitor system resources during operations. - Parallel Tasks: Enjoy the benefits of threaded execution. - Preset Saving: Save your quantization presets for future use. - iMatrix Generation: Effortlessly generate iMatrix. - Extensive Logging: Access detailed logs for better tracking and troubleshooting. 🍦 𝗖𝗼𝗺𝗽𝗮𝘁𝗶𝗯𝗶𝗹𝗶𝘁𝘆 & 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆: - Cross-Platform: Compatible with multiple platforms. - Open Source: Licensed under Apache-2.0. - Language Support: Available in 28 languages. - Executable Releases: Windows and Ubuntu users can download the latest release executable built with PyInstaller for enhanced performance, while other platforms can run it from source. 🕸️ 𝗦𝗶𝗺𝗽𝗹𝗶𝗳𝗶𝗲𝗱 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄: AutoGGUF removes the need for command line operations, automates directory creation, and offers extensive customization options. It addresses common pain points in the quantization workflow, making it easier to work with GGUF models. 👁️ 𝗚𝗶𝘁𝗛𝘂𝗯 𝗥𝗲𝗽𝗼𝘀𝗶𝘁𝗼𝗿𝘆: https://lnkd.in/dqTzAShC 🧠 𝗞𝗻𝗼𝘄𝗻 𝗜𝘀𝘀𝘂𝗲𝘀: - Preset Saving During Quantization: May cause UI thread crashes. - Task Deletion During Processing: You must cancel the task first to avoid crashes. 🤖𝗨𝗽𝗰𝗼𝗺𝗶𝗻𝗴 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀: - Custom command line parameters - Additional iMatrix generation parameters - Perplexity testing - Conversion of HF safetensors to GGUF - Enhanced progress tracking Below a screenshot of the app to give you a glimpse:

2

Manifold Learning via Foliations and Knowledge Transfer by Eliot Tron and Rita Fioresi https://lnkd.in/eK4McXUf Abstract Understanding how real data is distributed in high dimensional spaces is the key to many tasks in machine learning. We want to provide a natural geometric structure on the space of data employing a deep ReLU neural network trained as a classifier. Through the data information matrix (DIM), a variation of the Fisher information matrix, the model will discern a singular foliation structure on the space of data. We show that the singular points of such foliation are contained in a measure zero set, and that a local regular foliation exists almost everywhere. Experiments show that the data is correlated with leaves of such foliation. Moreover we show the potential of our approach for knowledge transfer by analyzing the spectrum of the DIM to measure distances between datasets.

96

1 commento