About
As a data person with know-how in analytics, data engineering and machine learning, I…
Activity
Experience
-
-
-
-
-
Education
Other similar profiles
Explore more posts
-
Sudarshan Lamkhede
The 3rd edition of Personalization and Recommendations in Search (PaRiS) workshop is tomorrow at SIGIR. We have an exciting array of speakers and panelists as well as a couple of submitted papers. 9:15 am EDT - Michael Bendersky - Personalized Generation: Research Challenges & Opportunities 10:30 am EDT - Negin Rahimi - Toward Interpretable Information Access Systems 11:30 am EDT - Alane Suhr - Interactive Language Agents: Training, Evaluation, and Interface 1:00 pm EDT - Anne Cocos - More Like This…how? Combining engagement and semantic signals for recommending similar titles at Netflix 1:50 pm EDT - Surya Kallumadi - Task aware personalization in eCommerce search and recommendations 2:20 pm EDT - Tejaswi Tenneti - Improving Relevance and Product Discoverability in Grocery search results 3 pm EDT - Paper Presentation 4 pm EDT - Panel discussion on "The Role of Generative AI in Revolutionizing Personalization, Recommendations, and Search". See you all there! PaRiS Organizers (Moumita Bhattacharya, Hongning Wang, Hamed Zamani, and your's truly :) ) #sigir2024 #research #genai #llm #search #recommendersystems #personalization
5 Comments -
Elena Sokolova, PhD
Want to give a talk about Managing ML project at the largest ML conference in the world? Submit your talk proposal to the KDD Workshop on Applied ML Management. The workshop focuses on real use-cases and take-aways from managing machine learning projects, teams, and products. It is a great opportunity to engage in discussions with industry leaders. More info is here: https://lnkd.in/ewrxKniT See you at KDD-2024 in Barcelona! #mlmanagement #kdd #amazonscience #conferences #ml #ai #science
1 Comment -
Amrita Rath
LLM Model Evaluation Metrics - When and How to Use Them The rapid evolution of Language Model (LLM) applications presents a dynamic landscape for evaluating model performance. How do you navigate this ever-changing field to ensure your models meet the highest standards of effectiveness and reliability? For evaluating LLMs, how do you discern which ones are most suitable for your specific NLP tasks? Despite the progress made, challenges persist in accurately assessing LLM performance. What are some of the biggest hurdles you've encountered in evaluating large language models, and how have you overcome them? What are some best practices you follow when evaluating LLMs to ensure robustness and reliability?
17 Comments -
David Bauch
Hi there, it’s David, Logan, and Sabrina! We’re building Borea (https://borea.dev), an open-source, self-hosted alternative to Amplitude, Heap, and Mixpanel (among others). Current product analytics software is great, but most are rigid black boxes that don't let you control your own user's data and are full of bloated, confusing, and unused features that don’t answer the big question: how are users using my product? So we worked to fix that with Borea. Borea enables front-end event capture and session recording with a simple dashboard of insights that you can understand. You can also utilize heat maps and set custom events with our API or embedded toolbar. We started with the goal of making gathering feedback more efficient so that developers could focus on building better products. Our first product was a mobile A/B testing platform that would allow developers to gain direct feedback on their mobile apps with a small group of beta users without having to force users to update their apps. Cool, right? As we began building, we realized this was way outside our technical wheelhouse. However, we did notice that the analytics side lacked solutions that were built to be simple, understandable, and privacy-minded. So we pivoted to build Borea: product analytics built for devs by devs. We currently use a custom, streamlined version of an older PostHog release for data processing, while we build our own tracking infrastructure, including SDKs, tracking scripts, and cloud deployment. Borea is built to work on any platform and stack. We have a simple install process and are currently building a JavaScript library for more fine-tuned event capturing. But until then, we have an API for that. Give it a try: <github-repo-link>. Share any feedback with us here, or on our Discord channel: https://lnkd.in/gWyVf7tU
-
Anoop Maurya
Unlocking the Power of Optical Character Recognition with Ollama OCR In today’s digital world, extracting text from images efficiently is crucial. Enter Ollama OCR! 🚀 An open-source Optical Character Recognition (OCR) application that uses advanced vision language models to transform how we interact with text in images. Inspired by LlamaOCR, Ollama OCR combines cutting-edge technology with a user-friendly interface. ------ Why Choose Ollama OCR? 🤔 ------ High-Precision Text Recognition 🎯 LLaVA 7B: Fast and efficient for real-time processing. Llama 3.2 Vision: High accuracy for complex documents. These models ensure reliable text extraction from various image types!Versatile ------ Output Formats 📄 Ollama OCR supports multiple output formats:Markdown: Preserves formatting with headers and lists. Plain Text: Clean and simple extraction. JSON: Structured data for easy integration. Structured Data: Extracts tables and organizes information. Key-Value Pairs: Efficiently extracts labeled information. ------- User-Friendly Interface 🖥️ Key features include:Drag-and-Drop Upload: Simplifies image addition. Real-Time Processing: Instant results enhance workflow. Download Options: Save extracted text in various formats. Image Preview: Review images before extraction. Technical Excellence 💪 -------- Ollama OCR is built on a robust architecture using Streamlit for the frontend and Python for the backend, ensuring smooth operation and reliable performance. Comprehensive error handling enhances user experience.Join the Community 🌍 -------- As an open-source and free tool, Ollama OCR welcomes contributions from developers and enthusiasts. Your involvement can help shape the future of OCR technology! Transform your text extraction experience today! Check out Ollama OCR on GitHub: https://lnkd.in/grnN_yMk
-
Shivendra Upadhyay
The idea of agents has been around for a while. What’s new here? In his talk, Andrew Ng explains agentic systems that use design patterns such as reflection, tool use, planning, and multi-agent collaboration to outperform the results of zero-shot prompting. #llms #aiagent #ml #rlhf #mlops #datascience #llmops https://lnkd.in/d2Q8Yj7X
-
Saravana Kumar
GenAI Meets ClassicalML at Apollo.io For the past year, the team at Apollo has been heads down developing lead scoring and search personalization models based on our users' prospecting activity. Recently our product team challenged us to create such models right at signup, even before any substantial activity had taken place. They presented a compelling vision of what they wanted to achieve. While we've integrated GenAI into numerous features and experiences, we were initially hesitant to apply it to Search and Scoring due to the sheer scale. Imagine leveraging even the most modest LLMs on over 100 million records—the latency and cost are significant hurdles, despite the promising capabilities of LLMs (which we're very excited about for the near future!). One innovative idea we explored was using GenAI to generate classical model definitions for some of these use cases. From the start, the results were promising but not without challenges. We had to use agents to gather comprehensive information about our users that could serve as a viable alternative to their activity data. Thankfully, with agent frameworks and our in-house company information data, we had a head start. Next was infusing our proprietary prospecting pattern data into these models and this is where Apollo’s rich historical data and fine-tuning helped address this. Lastly, getting GenAI to confirm schemas is where the newer LLM frameworks are helping enforce this much better. Kudos to Anurag Bajpai and Tyler Phillips for pursuing this idea, and we're thrilled that our early beta launch has been incredibly successful. Proud to be part of a team that pushes beyond the obvious, making new strides in engineering and product impact. Excited about the unique ways GenAI will continue to shape the future of ML.
1 Comment -
Hoàng Vũ
It's been a while since my last post, and now I'm back with a more thrilling concept in A/B Testing that will transform how we interpret and leverage data for decision-making. The main points of this post will focus on: - Correlation vs Causation: What if our A/B tests are only showing correlations but not the true causal impact of changes? Understanding the difference is crucial for making effective data-driven decisions. - Simpson's Paradox in A/B Testing: How often do we misinterpret A/B test results due to Simpson's Paradox? This paradox can obscure true relationships within data, leading to misleading conclusions. - Applying Causal Inference Approach through Uplifting Model Technique: How can we leverage causal inference to improve our A/B testing? By using the uplifting method, we can identify which segments of our audience are most influenced by interventions, moving beyond average treatment effects to uncover the true impact. You can read in details at: - Medium: https://lnkd.in/gvAfuq_d - Substack: https://lnkd.in/gqrh5MSD #DataScience #ABTesting #CausalInference #UpliftModeling #DataDrivenDecisions
-
Volker Janz
𝐓𝐨 𝐚𝐥𝐥 𝐝𝐚𝐭𝐚 𝐞𝐧𝐭𝐡𝐮𝐬𝐢𝐚𝐬𝐭𝐬: I highly recommend checking out the Data Engineering and Machine Learning Summit (DEML), happening on October 3rd and 4th this year! Hosted by Seattle Data Guy, co-hosted by Data Engineer Things, this free-to-attend 🫶 online conference features talks from those who truly get their hands dirty with real-world data 💡! 📚 Tracks: - Data Engineering & Data Infrastructure - Data Science & Machine Learning - Tech Demo Showcases ✨ Announced speakers (CFP is open): - Jun He: Maestro - Netflix’s data orchestrator - Himateja Madala: Data lineage and metadata management - Shachar Meir: Stand out in data interviews Check out the post by Xinran Waibel for the CFP form and registration link. It is awesome to see such a great event provided for free, emphasizing the importance of knowledge sharing and supporting each other in the (data) community 💚. Feel free to share, and if you're interested in attending, speaking, or sponsoring the conference, check out the linked post for more details! #dataengineering #softwareengineering #machinelearning #ai #datascience #dataconference #knowledgesharing
1 Comment -
Ilia Larchenko
I have just released an update to my AI Interviewer service (https://lnkd.in/gE74X6eA) that I presented last week. 🚀 Faster Interactions: All models are now in streaming/chunk processing mode. This improvement has drastically reduced response times, allowing for long answers and solutions to be processed quickly. GPT-4 works smoother now, and the online demo version is fast enough for practical use (but the quality of the locally running service is still much better). 🛠️ More Interview Types: Alongside the coding interview, I’ve added beta versions of five new types of technical interviews: - System Design - Machine Learning System Design - Machine Learning Theory - Math, Stats, and Logic - SQL Everything is still in the testing mode and there’s plenty of room to improve, but now you can simulate nearly any popular type of technical interview. What other types of interviews should I add in the future? Are there types of interviews you think LLMs can't handle yet? I’d love to hear your thoughts! #ai #chatGPT #interview #coding #automation #systemdesign #machinelearning #interviewpreparation
1 Comment -
Baran Köseoğlu
Hi network, I’ll be speaking at PyData Global 2024 alongside Egor Kraev, our data science lead at Wise! 🎉🎉 I’m thrilled to invite you to our session titled “Fast, intuitive feature selection via regression on Shapley values” at PyData Global 2024. This incredible event brings together some of the brightest minds in data science, machine learning, and beyond, and we’re honored to be a part of it! 📅 When: 04-12-2024, 14:30–15:00 (UTC) 📍 Where: Online – accessible from anywhere in the world! In our talk, we’ll explore our feature selection framework, shap-select. For a deeper dive, check out our paper and our repo here: https://lnkd.in/e6PYcqTJ, https://lnkd.in/e4Vsnu3b. Whether you’re a product manager, data science enthusiast, machine learning practitioner, or simply curious about feature selection, this session will have something for you. We can’t wait to share our learnings and engage in discussions with all of you! 👉 Register here to attend live: https://lnkd.in/dZkJ_FAM. Let’s connect, learn, and grow together at PyData Global 2024! See you there!🙌 #PyDataGlobal #DataScience #MachineLearning #FeatureSelection
-
Patryk Binkowski
📢 DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads🚀 In this work, they tackle the significant memory and computational challenges associated with deploying long-context large language models (LLMs). The proposed framework, DuoAttention, introduces a novel distinction between Retrieval Heads and Streaming Heads to optimize the use of the KV cache, reducing memory usage by up to 2.55× and improving decoding speeds by up to 2.18×, all while maintaining high accuracy. Technical Highlights: 1. Retrieval Heads vs. Streaming Heads: DuoAttention divides attention heads into two categories. Retrieval Heads are crucial for processing long contexts, whereas Streaming Heads focus on recent tokens, allowing for a more efficient use of memory resources. This division helps in reducing the computational load without compromising model accuracy. 2. KV Cache Optimizatio: The framework uses selective caching strategies, applying full KV cache to Retrieval Heads while maintaining a constant-length cache for Streaming Heads. This results in significant reductions in both memory usage and latency during long-context inference. 3. Scalable Long-Context Inference: With DuoAttention, models like Llama-3-8B can handle up to 3.3 million tokens on a single A100 GPU. This is made possible through careful optimization of memory allocation and the use of lightweight algorithms to identify the critical heads. 4. Performance Improvements: DuoAttention achieves up to 2.55× memory reduction for multi-head attention (MHA) models and up to 1.67× for grouped-query attention (GQA) models. It also speeds up decoding by up to 2.18× for MHA models and up to 1.50× for GQA models, providing significant benefits for real-world applications involving large language models. 5. Compatibility with Quantization: The method is fully compatible with quantization techniques, such as 8-bit weight and 4-bit KV cache quantization. This compatibility further boosts efficiency, enabling the model to handle larger contexts with reduced hardware requirements. With DuoAttention, the team has achieved long-context capabilities for models like Llama-3-8B, enabling handling of up to 3.3 million tokens on a single A100 GPU. This advancement marks a significant step towards more efficient, scalable LLM deployments! 🔍 Paper: https://lnkd.in/dDFHimSA #LLMs #MachineLearning # #EfficientInference
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Leo K. in United States
-
Leo K
CyberSecurity Subject Matter Expert | Master of Engineering | CISSP | CASP | | CISA | CEH | CompTIA Security+
-
Leo K
AI Innovator
-
Leo K
--
-
Leo K.
67 others named Leo K. in United States are on LinkedIn
See others named Leo K.