Mehdi Ouazza

#ObsessivelyEnthusiasticInvestor #UnreasonablyEnthusiasticInvestor #IrrationallyEnthusiasticInvestor at REFASHIOND Ventures: The Industrial Transformation Fund. We look for the people we believe are doing the best work in Industrial Transformation & Supply Chain Technology; Attempting the impossible. Then we go all-in on them. With obsessive and unreasonable enthusiasm. Join us. #IndustrialTransformation #SupplyChain #Innovation #Technology #Startups #VentureCapital #DemandTheImpossible #UnreasonableEnthusiasm #ObsessiveEnthusiasm #IrrationalEnthusiasm 

6

There's a big difference between Database Modeling and Semantic Layers- How would you explain it? Even though they look alike, the underlying meaning is vastly different. ▶ The type of data modeling you would do with database modelers like SqlDBM and semantic layers are interrelated in that we help design and organize database structures in a way that supports the creation of a semantic layer. However: ▶ A database modeling solution is focused on the 𝐩𝐡𝐲𝐬𝐢𝐜𝐚𝐥 and 𝐥𝐨𝐠𝐢𝐜𝐚𝐥 aspects of database modeling, ensuring the data structures align with technical requirements and business concepts. ▶ A semantic layer operates at a higher level of abstraction often within 𝐁𝐈 𝐭𝐨𝐨𝐥𝐬 such as PowerBI, Looker, ThoughtSpot, and Tableau. It maps the technical details of data that were designed by tools like SqlDBM or not properly designed at all, to business-friendly terms and concepts. ▶A tool like SqlDBM will essentially lay the 𝐟𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐝𝐚𝐭𝐚 𝐚𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞 that a semantic layer translates into an easily navigable and understandable format for end-users. ---- "Enabling non-technical business users to perform data analysis without needing to understand complex database queries or structures. However, leveraging a data modeling tool and having a good data model will make the complexity of building and running a semantic layer more manageable and efficient." - Keith Belanger, 𝘚𝘲𝘭𝘋𝘉𝘔 𝘗𝘳𝘰𝘥𝘶𝘤𝘵 𝘌𝘷𝘢𝘯𝘨𝘦𝘭𝘪𝘴𝘵 & 𝘚𝘯𝘰𝘸𝘧𝘭𝘢𝘬𝘦 𝘋𝘢𝘵𝘢 𝘚𝘶𝘱𝘦𝘳𝘩𝘦𝘳𝘰 ---- This is why you hear someone say "We do our data modeling in Power BI" - they are modeling their 𝐬𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐥𝐚𝐲𝐞𝐫 in Power BI. This is an image from Power BI, if it had Dark Mode it would look even more similar! #DataModeling #DataArchitecture #SqlDBM

55

1 Kommentar

Really excited about the 'Hudi-rs' release 🎉 Not every data workload requires the need for distributed processing and to setup Spark & JVM to execute them. In fact, there has been an increasing trend in terms of using Rust & Python-based compute engines for a range of workloads. The hudi-rs project is a native Rust library for Apache Hudi, with bindings into Python. This opens up an exciting new world where users can read Hudi tables using non-JVM compute engines. Under the hood, hudi-rs uses #ApacheArrow. So, it scans the Hudi table and reads out records as Arrow RecordBatch. RecordBatch is a table-like data structure that is semantically a sequence of fields, each a contiguous Arrow Array. Once the records are in an Arrow Table, you can easily use it with Arrow-native and integrated tools such as DuckDB, Polars, Daft etc. for further analysis. This is an exciting new direction for open lakehouse platforms like Hudi that brings differentiating capabilities such as: incremental processing, multi-modal indexing subsystem, open table services, among others. We would also love to have any sort of community contributions & PRs. Check out the contributing link & repo in comments. #dataengineering #softwareengineering

127

3 Kommentare

Leakage is the silent killer of predictive models! If you're a data practitioner taking your first steps in ML and predictions, learn how to identify and prevent leakage below 👇 ‎

59

7 Kommentare

Kolmogorov–Arnold Networks seem like a paradigm shift in ML architectures. Curious to see how they evolve. "While MLPs have fixed activation functions on nodes ("neurons"), KANs have learnable activation functions on edges ("weights"). KANs have no linear weights at all -- every weight parameter is replaced by a univariate function parametrized as a spline. " https://lnkd.in/gdPk_qCQ

29

I'm very happy to share a technical article I co-created with Deeksha Gunde, that walks through enabling a federated Airflow environment using AWS SNS/SQS. https://lnkd.in/e4RwPirp #aws #dataengineering #analytics #bigdata

33

Exciting #news for the #generativeai field! Today #antrophic announced: 👉 an upgraded Claude 3.5 Sonnet. 👉 a new model: Claude 3.5 #haiku which matches the performance of Claude 3 Opus on many evaluations for the same cost and similar speed to the previous generation of Haiku. 👏 a groundbreaking #computeruse capability which allows people to direct Claude to use computers the way we do: i.e. by looking at a screen, moving a cursor, clicking buttons, and typing text. Whilst at this stage this is still experimental, it is great to see this new capability being released early for #feedback from developers to further improve the model's performance. If you have a couple of minutes, have a look at the #computeruse demo: it is truly impressive!

5

IMVC 2024 just released Dr. Amir Alush great talk about going from messy visual data to high quality datasets: https://lnkd.in/dGBDUUHQ . Try it yourself on your own data at https://lnkd.in/ds4vwrpP!

24

🛠️ RAG On-Prem. Built on Kubernetes, fine-tuned open source models, and battle-tested on 2B+ tokens. 🧿 GroundX has made waves with clients like Air France and Dartmouth, and so far, we’ve processed over 2 billion tokens of data! And now, with Enterprise Grade RAG On Prem— it's designed from the ground up for secure, closed environments. Our API's offer the same high-performance RAG (Retrieval-Augmented Generation) capabilities but fully deployable behind your firewall. We built this for organizations where security isn’t optional but essential—think finance, healthcare, and defense. Are you exploring AI solutions for secure environments? Let’s talk about how GroundX On Prem can meet the most demanding requirements. Learn More: https://buff.ly/2DiomCf #EnterpriseAI #RAG #kubernetes #onprem

5

Dask DataFrame is fast now - Florian Jetter (Coiled) @ PyData Südwest Live stream of the PyData Südwest meetup https://lnkd.in/eU9G8rvu 20:15 Talk / Q&A 20:45 Talk Lightning Talks Ask questions via Slido: https://bit.ly/SW2405QA 📺 Dask DataFrame is fast now - Florian Jetter (Coiled) Dask is a library for distributed computing with Python that integrates tightly with pandas. Historically, Dask was the easiest choice to use (it's just pandas) but struggled to achieve robust performance (there were many ways to accidentally perform poorly). The re-implementation of the DataFrame API addresses all of the pain points that users ran into. We will look into how Dask is a lot faster now, how it performs on benchmarks that is struggled with in the past and how it compares to other tools like Spark, DuckDB and Polars. ⚡️ Lightning Talks

28

2 Kommentare

This week I'm considering the benefits of so-called PostModern Data tooling (inspired by Tomasz Tunguz's recent post). Why is it such a good model to start with OSS and then offer a paid version? Shoutout to Dagster Labs, MotherDuck, Tobiko, ClickHouse, Lightdash and of course Cube, who I count among these tools. https://lnkd.in/e-nj_XUS

13

AWS has optimized the PyTorch torch.compile feature for AWS Graviton3 processors, resulting in up to 2x better performance for Hugging Face model inference and up to 1.35x better performance for TorchBench model inference 🔥 Read more in our latest blog: https://hubs.la/Q02FRNQt0

119

2 Kommentare

Build an MLP (multi-layer perceptrons) When building an MLP: Consider that an MLP can be used to solve multiple types of machine learning tasks, including classification and regression. Consider using neural networks when you have extremely large datasets—datasets with upward of hundreds of thousands, and even millions of data examples. Scale your dataset's features before training them on the network. Consider selecting the tanh or rectified linear unit (ReLU) activation functions for the hidden layer(s). Consider using the sigmoid activation function for the output layer of a classification problem. Consider that a ReLU variant like leaky ReLU can help overcome the vanishing gradients problem. Consider starting with just a couple hidden layers, as this will be sufficient for many types of tasks and will require much less training time. For complex tasks where time is less of a concern, consider gradually increasing the number of hidden layers until you start to overfit the training data. Consider building each hidden layer with the same or similar number of neurons. As with the number of hidden layers, start with a small number of neurons per layer and then gradually increase that number until the network starts overfitting. Use Python for MLPs The scikit-learn MLPClassifier() and MLPRegressor() classes enable you to construct an MLP neural network. The following are some of the objects and functions you can use to build such a network. model = sklearn.neural_network.MLPClassifier(hidden_layer_sizes = (5, 5, 5), activation = 'relu') —This constructs an MLP to use for classification. In this case, three hidden layers will be constructed, each with five neurons. The ReLU activation function is being applied to the hidden layer neurons. model = sklearn.neural_network.MLPRegressor(hidden_layer_sizes = (2, 2), activation = 'tanh') —This constructs an MLP to use for regression. In this case, two hidden layers will be constructed, each with two neurons. The tanh activation function is being applied to the hidden layer neurons. You can use these class objects to call the same fit(), score(), predict(), and predict_proba() (classification only) methods as before, as well as any of the applicable metrics methods. model.coefs_ —An attribute that returns the weights between the neurons of each layer. model.intercepts_ —An attribute that returns the bias terms for each layer. model.loss_curve_ —An attribute that returns the change in loss values for each iteration of a gradient descent solver. Source: CAIP.

2

A Must watch for all RAG enthusiasts. A gem of video vy Qdrant. Optimizing ColPali for Retrieval at Scale, from Theory to Practice Main Takeaways: 1. Visual RAG 2. Vector Quantization 3. Some business use cases. Watch it here: https://lnkd.in/e-dYEfyF

20

1 Kommentar

In Daniel Klitzke's latest article, he shows how you can use the Huggingface Transformers and Sentence Transformers libraries to boost you RAG pipelines using reranking models. #LLM #MachineLearning

23

int8 embeddings from Cohere (4x saving in memory, 30% latency for retrieval, 0.01% loss in search quality (yes, it's 10^{-4}) ) binary embeddings (yes, it's 1 bit, 32x saving in memory), led to 6-10% percent loss in search quality measured by MTEB that might be reasonable for some applications, where 5% retrieval noise is critical or might be reduced through other means there are technique how to improve it See details inside https://lnkd.in/gzEHh-Wq

22

🌟 Live RAG Comparison Test: Pinecone vs. MongoDB vs. Postgres vs. SingleStore ❓Why It Matters: 🔹The role of Retrieval-Augmented Generation (RAG) has become increasingly significant in today's #GenAI landscape. RAG combines the best of both worlds of generative and retrieval-based models to provide more contextually accurate and rich responses in LLM apps.  🔹Latency is crucial for RAG in LLM apps because it directly impacts the user experience by determining how quickly and interactively the model can retrieve and integrate relevant information to generate accurate responses 🔹Rohit Bhamidipati will offer a hands-on RAG performance evaluation of leading back-end/database vendors such as Pinecone vs. MongoDB vs. Postgres vs. SingleStore, comparing throughput, efficiency, and latency for large datasets. 📚 What You'll Learn: ➡️ The mechanics of RAG and its impact on enhancing language model responses. ➡️ How Vector Databases like Pinecone, MongoDB, PostgreSQL, and SingleStore facilitate the functionality of RAG. ➡️ Comparative analysis showcasing real-time performance metrics of these databases. ➡️ Best practices for integrating these technologies into your AI and ML projects to boost efficiency and accuracy. ⚡️ Can't make it? No worries! All registrants will receive a copy of the webinar recording and additional resources via email post-session ✅ Live demo and code-share session will be offered for a hands-on experience 📅 Event Details: 🔹 Thursday, May 9 @ 10-11 AM PDT 🔹 #Free Registration: https://lnkd.in/g6EvB8NG #artificialintelligence #rag #ad

19

1 Kommentar