Kris Dorosz

This will unlock vector search for the 10-100x number of use cases where it has been too expensive until now. A must read if you're using embeddings.

5

The Machine Learning for Irregular Time Series (ML4ITS2024): Advances in Generative Models, Global Models and Self-Supervised Learning Workshop event located at the ECML PKDD 2024 conference, and will take place on September 9-13, 2024, in Vilnius, Lithuania. The main scope of this workshop is to advance the state-of-the-art in time series analysis for “irregular” time series. We define time series to be “irregular” if they fall under one or several of the following categories: Short: univariate and multivariate time series with a limited amount of data and history; Multiresolution: multivariate time series where each signal has a different granularity or resolution in terms of sampling frequency; Noisy: univariate/multivariate time series with some additional perturbation appearing in different forms. In this class, we also include time series with missing data; Heterogeneous: multivariate time series, usually collected by many physical systems, that exhibit different types of embedded, statistical patterns and behaviours; Scarcely labeled and unlabeled: univariate/multivariate time series where only a small part of the data is labeled or completely unlabeled. workshop include, but are not limited to: Generative models for Synthetic Data generation, including GANs, diffusion models and masked modeling in time series domain, Explainable AI techniques tailored to deep time series models, Uncertainty quantification in deep time series models, Methods for Data Imputation and Denoising, Transfer Learning for Time Series forecasting and classification, including FNN, CNN, and Recurrent NN, Transformers architectures and Attention mechanisms for Time Series analysis, Graph Neural Networks for Anomaly Detection and Failure Prediction, Organizers: Massimiliano Ruocco (SINTEF Digital / Norwegian University of Science and Technology) Erlend Aune (HANCE / Norwegian University of Science and Technology) Claudio Gallicchio (University of Pisa) More: https://lnkd.in/epc86BBE

21

Happening July 11 at Arize:Observe: Sandeep Subramanian from Mistral AI will be talking about CodeStral–LLMs for code... This talk will cover some of the important aspects of building LLMs for code. Specifically, it will discuss the ingredients of CodeStral-22B, a code model from MistralAI that is trained with FIM, allowing it to act as a "tab-complete" model as well as a chat model. The Research Track at Arize:Observe covers cutting-edge research, emerging techniques, and theoretical advancements in AI observability and LLM evaluation. Explore the latest findings and methodologies pushing the boundaries of what’s possible in understanding, evaluating, and improving AI-powered systems. Join us SHACK15 on July 11 in SF for latest cutting-edge research, real-world insights from AI engineers, and useful workshops to evaluate and troubleshoot AI in production. See more speakers and register: https://lnkd.in/gpEitQb5

68

6 Comments

Curious how we got into the underlying mechanisms of the Ragas framework for synthetic test set generation and RAG evaluation? ⁉️ Luka Panić presents a practical demonstration of its integration with LangSmith, a powerful monitoring platform for AI language systems, that helps demystify Ragas processes. 📝🔥 The new episode of our AI blog series sure is worth reading ➡️ bit.ly/3RmgEuk #PixionAiBlogSeries #AI #LLM #RAG

9

2 Comments

🔦 Spotlight on Dean Shaff: Building our own Async Event Loop in Python At Yukka Lab we are working with complex technologies. After all, we are not just analyzing news, figuring out the tone or sentiment in the news, and analyzing what it means and how it relates to the company; we also need to make sure our system works efficiently and in real-time. That means we have to understand the systems we work with in-depth so we can use them in the most efficient way possible. Dean's latest blog, "Building our own Async Event Loop in Python," perfectly illustrates this commitment.  In this insightful piece, he demystifies Python’s async model by constructing his own version of “asyncio.sleep”. 🔗 Discover how we push the boundaries of technological innovation to drive better results. Read the full article here: https://lnkd.in/eYCUR7NZ #YukkaLab #PythonProgramming #Innovation #TechInsights #DataProcessing

25

1 Comment

Exciting news from Mistral AI! At the end of last week Mistral AI released a new version of their small 7B parameter model. The new model has an extended vocabulary and now supports function calling. In addition, they also dropped a fine-tuning cookbook to accompany this model (and their other ones). Fine-tuning is a great technique to use for narrow tasks, such as summarisation or building sections of large documents. Fine-tuning allows a business to “alter” the weights of a model to be more specific to their dataset. In my opinion, Mistral AI continue to provide valuable models and research into the Generative AI industry. They’re not like OpenAI and Anthropic who are on an ambitious (and expensive) journey to build an “AGI” model. Mistral AI are building useful models for businesses. The models they release are easily portable so companies can use them in their own secure environment. Moreover, the models are cost effective to both fine-tune and run in production. Here is a link to the fine-tuning repo: https://lnkd.in/e4xaPjKf

13

If you are interested in building a deep network for point clouds but don't know where to start, come to our CVPR tutorial: All You Need to Know about Point Cloud Understanding! We will discuss many architectures and design choices as well as code-level optimizations! The tutorial will happen on Tuesday morning starting at 9AM at Summit 444 https://lnkd.in/gR3ExTy5 Hengshuang Zhao Xiaoyang Wu Zhijian Liu

65

In the CUQI project we study, implement, and utilize the Bayesian framework for inverse problems. A recent paper co-authored by CUQI team member Mirza Karamehmedovic employs this framework for the inference of the trajectory and intensity of a source in subsonic motion, based on measurements of the radiated field. Specifically, the authors can incorporate prior information on the source trajectory. Read more in Wang, S., Karamehmedović, M., and Triki, F., 2023. Localization of moving sources: uniqueness, stability, and Bayesian inference. SIAM Journal on Applied Mathematics, 83(3), pp. 1049–1073. CUQI_DTU is supported by Villum Fonden and DTU Compute #appliedmathematics #scientificcomputing

12

1 Comment

#IJCAIchallenge OPEN CHALLENGE ⏰🏁 Financial Challenges in Large Language Models - FinLLM at #IJCAI2024 ➡️https://lnkd.in/d7StmYwx With the advent of LLM in finance, financial text analysis, generation, and decision-making tasks have received growing attention, including financial classification, financial text summarization, and single stock trading. For instance, 1) The financial classification task aims to categorize sentences as claims or premises effectively. 2) Financial text summarization task for abstracting financial texts into concise summaries. 3) The single stock trading task aims to make a predictable decision regarding stock trading. Although several approaches have achieved remarkable performance with LLMs, their capabilities of comprehensive analysis and decision-making for finance remain largely unexplored. We plan to explore the ability of LLMs from these facets. We prepared a Large Language Model-based financial shared task, FinLLM, which contains three subtasks to address diverse financial challenges effectively and holistically.  

1

AnchorGT: A Novel Attention Architecture for Graph Transformers as a Flexible Building Block to Improve the Scalability of a Wide Range of Graph Transformer Models Transformers have taken the machine learning world by storm with their powerful self-attention mechanism, achieving state-of-the-art results in areas like natural language processing and computer vision. However, when it came to graph data, which is... https://lnkd.in/eedeAnTa #AI #ML #Automation

2

In Kraina.AI we are training a large transformer model on spatial data, in order to be able to do it, we need to be able to have in memory storage of spatial data. So we had to build a library that handles that. Kamil Raczycki wrote it - quackosm - and recently he explained why it is so important in spatial transformer training on the wonderful podcast/vidcast of Matt Forrest. Enjoy! There's a lot of use cases that need to have in memory spatial data - feel free to share this link so that people know we've released any easy solution for a common engineering task. CC: Thomas Wolf this is what I wrote to you about :) https://lnkd.in/dv2n2B8f

19

🚀 How's the new CodeMistral model? 🤔 And which code model should you choose to fine-tune for your specific tasks? We present `PR-Agent fine-tuning benchmark`. It methodically compares various open-source models based on their fine-tuning capabilities. Check it out now! 👇 https://lnkd.in/e7-vzAMt

25

2 Comments

PyMilo 0.8 is out now! 🚀 You can now transport your #scikitlearn ensemble models. Check out your favorite decision tree ensemble model weights. Previous updates on PyMilo: - Version 0.6 with the main focus on SVM models support - Version 0.7 with the main focus on Nearest Neighbors support To find more, go to https://lnkd.in/dCgu2VeD #Python #MachineLearning #opensource #DataScientists

9

1 Comment

Just finished the course “Getting Hands-on with GPT-4: Tips and Tricks” by Denys Linkov! Check it out: https://lnkd.in/dnPPMFtd #generativeai #gpt4.

1

I ❤️ TensorFlow. It’s one of the most versatile solution in developing and deploying deep learning models in the real-world. Here are 15 operations within TensorFlow I use often in my deep learning workflows👇 1. tf.keras.layers.Dense() # The core building block for NN 2. tf.keras.optimizers # Optimization (e.g., Adam, SGD) 3. tf.reduce_mean() / tf.reduce_sum() # Common math operations. 4. tf.data.Dataset # Create data pipelines for training 5. tf.GradientTape() # Automatic differentiation. 6. tf.keras.losses # Loss functions 7. tf.nn.softmax() # Essential for classification tasks 8. tf.keras.layers.Normalization() # Normalize continuous data 9. tf.cast() # Convert data types 10. tf.keras.layers.Dropout() # Reduces overfitting 11. tf.where() # Filter data 12. tf.argmax() / tf.argmin() # Common math operations 13. tf.reshape() # Reshape dimension of data 14. tf.keras.metrics # Model evaluation 15. tf.keras.layers.Embedding() # Creates embedding vectors Know these functions for your day-to-day jobs in ML engineering and data science (specialized in deep learning), and, for ML coding interview questions! 👉 What are your must-know TensorFlow (or PyTorch) operations? Drop a comment👇 👉 I will be teaching 𝗗𝗦 𝗜𝗻𝘁𝗲𝗿𝘃𝗶𝗲𝘄 𝗕𝗼𝗼𝘁𝗰𝗮𝗺𝗽 (https://lnkd.in/eiA5Ntdp) next week! Join the June cohort before it starts! Ace upcoming interviews with these👇 📕 𝗜𝗻𝘁𝗲𝗿𝘃𝗶𝗲𝘄 𝗣𝗿𝗲𝗽 𝗖𝗼𝘂𝗿𝘀𝗲𝘀: https://lnkd.in/gzgB-dHT 📙 𝗝𝗼𝗶𝗻 𝗠𝗟𝗘 𝗜𝗻𝘁𝗲𝗿𝘃𝗶𝗲𝘄 𝗕𝗼𝗼𝘁𝗰𝗮𝗺𝗽: https://lnkd.in/e6HbN6dy 📗 𝗔𝗕 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗖𝗼𝘂𝗿𝘀𝗲: https://lnkd.in/g82dMJ77 👉 Smash 👍 and follow Daniel Lee to land dream data & AI jobs 🚀

156

5 Comments

Dive into the capabilities of #Neo4j Vector Search! During this last livestream, David Pond and Alex focused on the exciting potential of vector indexes in Neo4j. ⬆️ These indexes facilitate efficient and accurate data searching by comparing the semantic similarities between vectors. https://bit.ly/4e3CkVM #vectorsearch #vector #Neo4j

19

Detecting hallucinations in large language models using semantic entropy Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn & Yarin Gal Abstract Large language model (LLM) systems, such as ChatGPT1or Gemini, can show impressive reasoning and question-answering capabilities but often ‘hallucinate’ false outputs and unsubstantiated answers. Answering unreliably or without the necessary information prevents adoption in diverse fields, with problems including fabrication of legal precedents5 or untrue facts in news articles and even posing a risk to human life in medical domains such as radiology. Encouraging truthfulness through supervision or reinforcement has been only partially successful8. Researchers need a general method for detecting hallucinations in LLMs that works even with new and unseen questions to which humans might not know the answer. Here we develop new methods grounded in statistics, proposing entropy-based uncertainty estimators for LLMs to detect a subset of hallucinations—confabulations—which are arbitrary and incorrect generations. Our method addresses the fact that one idea can be expressed in many ways by computing uncertainty at the level of meaning rather than specific sequences of words. Our method works across datasets and tasks without a priori knowledge of the task, requires no task-specific data and robustly generalizes to new tasks not seen before. By detecting when a prompt is likely to produce a confabulation, our method helps users understand when they must take extra care with LLMs and opens up new possibilities for using LLMs that are otherwise prevented by their unreliability. 👉 https://lnkd.in/dmgtiZ9A #machinelearning

2