Mike Golovan

Today we invite you to watch a recording of Contributed Talks from the ML in PL 2023 conference, where Błażej Osiński talked about: Toward Efficient Machine Learning For Sequential Decision Making. https://lnkd.in/datBvHb8 In this talk, I will present my personal journey toward developing efficient machine learning methods for sequential decision making. This journey spans both academia and industry, tackling diverse challenges that range from Atari games to autonomous driving and robotics. Why should we care about efficient methods for sequential decision making? If we observe the progress of artificial intelligence over the last few decades, it has managed to ascend what were perceived as pinnacles of human intelligence. From chess and Go to StarCraft and No Limit Texas Hold'em, machines have toppled human champions. The emergence of large language models extends this progress to more open-ended challenges. Despite these successes, when it comes to physical interactions with the real world, the achievements of AI are less remarkable. For instance, despite three decades of research, multi-fingered robotic hands are still far less dexterous than those of an average human child. The comparatively slower advancements in the realm of physical interactions can largely be attributed to the massive data requirements of contemporary machine learning methods. Advancements in efficiency could pave the way for truly practical robots capable of alleviating our household chores, rather than just beating us in games or eyeing our jobs. Błażej is a machine learning researcher with experience in sequential decision-making problems (reinforcement learning, imitation learning). He is about to graduate with a Ph.D. from the Uniwersytet Warszawski. He has divided his time between academia and industry, having affiliations with institutions such as Lyft Level 5, Google Brain, and the University of California, Berkeley. He is now pivoting to the ambitious endeavor of explaining cellular biology with machine learning working for Altos Labs.

4

1 коментар

Being a data engineer is both an art and a science—a blend of creativity, problem-solving, and technical wizardry. Having spent many years in the field, Andrey Torchilo knows firsthand what a typical day in the life of a data engineer looks like, and he shares his experience in his latest article. Read it here: https://hubs.ly/Q02Qn-n70

12

Hi, seems like I never posted a video of CVPR 2024 "Language meets driving" tutorial. Here it is. https://lnkd.in/eKHs65Ez - Why use Language models in end-2-end driving? - Measuring LLM capabilities in driving - Paper survey: Non-visual LLMs in self-driving - Paper survey: VLMs for perception in self-driving - Why is it important to go end-2-end with VLMs? - Paper survey: End-2-end VLMs in self-driving - Unsolved problems, conclusions, Q&A Also see older talks and publications on the new world-wide entertainment portal "sinavski.com" 😵 https://lnkd.in/eQuwyK-3 #LLM #AI #AutonomousCars #VLM #Robotics

71

4 коментарі

How to use a machine learning model on unstructured data? Last year our ML team had a chance to cooperate with a chatbot platform provider on building a knowledge base. The challenge was to use for it unstructured data, such as websites content.  However, because websites can be created in various ways, coming up with one universal set of rules for converting formatted text into data structures (parset) is difficult. 👉 Our approach was to consider only these parts of websites created with HTML5 semantics. It helps separate the valuable content from useless parts such as navbars, footers or pop-ups. 👉 Another step was to divide the vast amount of unstructured data into small chunks of information, such as paragraphs. 👉 Based on it, our team prepared an ML model for parsing fragmented data, which compared usually one-sentenced questions with 10, 15 or even 20-sentenced paragraphs.  The knowledge base powered by this machine learning model could show, for instance, five paragraphs that include an answer to a user’s question within them.     Check more details in our article linked in the comment section 👇 . #machinelearning #ML #knowledgebase 

8

1 коментар

More than 70% of computer vision developers struggle with sourcing training data. Join Jakub Pietrzak, CTO of SKY ENGINE AI, in a free webinar where he will describe how synthetic training data can solve this problem.  As an example, he’ll focus on how 3D generative AI combines with physically-based rendering to train deep learning models for the analysis of human faces, bodies and behavior. He will also demonstrate example datasets created in the SKY ENGINE AI Synthetic Data Cloud, along with inference results from models trained purely on synthetic data. https://lnkd.in/gk-eFcys 

9

Do you have an upcoming interview in data science? Alta (Zietsman) Saunders from Praelexis recently wrote a blog about how to survive such an interview. Read the blog here: https://lnkd.in/dfzPB3kp Remember, your soft skills are just as important as your technical skills! When considering your soft skills think about how you come across. When preparing for a data science interview, ask yourself the following questions: 1. Are you communicating clearly? 2. Can you communicate a complex idea (like defining a problem)? 3. How do you speak to the people in the interview? 4. Will the people who are interviewing you feel like you are a candidate that they can introduce to a client? 5. Are you able to see the bigger picture and communicate it? If you are curious about vacancies at Praelexis, contact us or email [email protected]. #praelexis #ai #datascience #interview #passion #tips #softskills

8

1 коментар

An enhanced N-BEATS model, N-BEATS, has been proposed by Mateusz Kasprzyk, Paweł Pełka, Boris N. Oreshkin, and Grzegorz Dudek for mid-term electricity load forecasting. This model builds on the strengths of the original N-BEATS architecture, introducing two key modifications: a novel loss function combining pinball and normalized MSE loss, and a modified block architecture with a destandardization component. Evaluated on real-world data from 35 European countries, N-BEATS demonstrated superior performance with lower MAPE, RMSE, and forecast error dispersion compared to other methods. This enhanced model offers a more efficient and less complex approach to mid-term electricity demand forecasting, showcasing the potential for improved accuracy and reliability in the field. https://lnkd.in/d3f4EvEi

3

Why is DataWalk perfect for managing large graphs that don't fit into RAM and for workflows requiring incremental data loading? Check out the benchmark study by our colleague Łukasz Łaszczuk https://lnkd.in/dCnuhqMF #knowledgegraph #benchmark

29

Ready to implement a custom Python processor for your AI data pipeline? Thanks to Weronika Falkowska, we just launched our newest tutorial on Datavolo DevCenter; Building a Python transform processor. This is the first in a series of Python development tutorials for Datavolo runtimes such as Datavolo Cloud. Try the hands-on tutorial for yourself at https://lnkd.in/gEHd4EzP #datavolo #python #nifi #dataengineering 

11

1 коментар

After exploring various retrieval strategies and Ragas as an evaluation tool, it's time to see how theory works in practice! 🛠️ In a new episode of our AI blog series, Stipan Petrović presents a case study of a scalable RAG application crafting that can be applied within the production environment of any real project at any time. ⚙️💡 AI things are getting real, don't miss out ➡️ bit.ly/3W3ZZ1f #PixionAiBlogSeries #AI #LLM #RAG

9

It wasn't so long ago that we were making waves as an MLOps Challenger in AIM Research's study, and now we're doubling down - Ganit has just been featured in the PeMa 2024 Data Engineering Quadrant! Ganit's capabilities are on the map, and we owe the maturity of our solutions to the customers that dare to dream data differently. And we aren't done disrupting yet - you know just who to call when you need that competitive edge.   See you in the Leaders Quadrant next time, AIM and AIM Research! Shivaprasad KT | Hariharan R. | Ashok Harwani | Dan Thorpe | Surya Pratap Singh | Karan Vichare | Sandeep M.Makwana #DataEngineering #DataInnovation #IndustryDisruptors #GameChangers #FeaturedChallenger #PeMaQuadrant2024 #TechTrailblazers #DataRevolutionaries #DreamDataDifferently #DataTransformation #FutureFirst #BigDataPioneers #AI #ML #Analytics #DataAnalytics #DataScience #ClientSuccess #Ganit #Ganitinc #DataSpeaks

74

1 коментар

Jelvix machine learning expert is here! 😎 Who is a machine learning engineer, and what are the requirements? We're going to cover all the important information in our newest video! Our colleague will explain his job through an example of the latest AI project he completed at Jelvix. He even shared his work schedule! ⬇️ Watch our video and don't forget to subscribe! #MachineLearning #ML #MLengineer

3

📢 Excited to share comparison on zero-shot-learning (ZSL) reasoning 🧠 in Sentiment Analysis with open models (Low Resource Domain, texts are in Russian). 😊 😢 😐 At the moment, 🔥 Mistral-7b-v2-instruct 🔥 is the best performed alternative to ChatGPT in author opiniong reasoning according to our studies that were recently higlighted and available on arxiv. Eventhougt, Mistral-7b-v2 outperforms ChatGPT-3.5-1106 (175B) by being 7B in nature! 🗒 Note (non-official): We separately tested the most recent chat versions 🔥 Mixtral-8x22B 🔥 outperforms and results in even closer capabilities of ChatGPT-4 in Reasoning Sentiment Analysis in Russian. Updates in the leaderboard @ Github below 👇 📜 Paper @ arxiv: https://lnkd.in/ew4tTfgN 📊 Leaderboard: https://lnkd.in/eRe_MSq7 📰 Dataset: https://lnkd.in/eSB2Wq8k Thanks everyone for interest and I believe these findings might be helpful in your studies that involve "reasoning in natural language", "multilingual stdudies" 🙏 #mistral #7b #mistral7b #reasoning #nlp #sentimentanalysis #benchmark

7

🔹Make the PySpark fly 🔹Magdalena Nowak Writing efficient Spark applications with Python can be surprisingly challenging. With PySpark library evolving fast many of the best practices aren't well-known, especially among Data Scientists. They tend to stick to the methods and workflows they know from more known libraries like Pandas, treating Spark code similarly - which often ends up being inefficient. Moreover, with most tutorials focusing on the basic features of the PySpark library, there is little knowledge about new concepts that make our lives easier and our data pipelines faster. This talk will help you progress beyond .groupBys and .counts. Watch the entire speech ➡️ https://lnkd.in/dJ_6UsxC

6

☝️ From tips 1-5, there’s an assumption that the model is fixed, but it gets even more complicated when you can change the model. On Day 6, we’re exploring model level optimizations, which include any method that changes the model or architecture itself. Since neural networks often contain many redundancies, it’s essential to exploit or streamline these redundancies to accelerate the inference process. What can you do to optimize inference at the model level?   𝐒𝐢𝐱𝐭𝐡 𝐭𝐢𝐩: 𝐂𝐨𝐦𝐩𝐫𝐞𝐬𝐬 𝐦𝐨𝐝𝐞𝐥𝐬 𝐰𝐢𝐭𝐡 𝐦𝐨𝐝𝐞𝐥 𝐝𝐢𝐬𝐭𝐢𝐥𝐥𝐚𝐭𝐢𝐨𝐧. 🧑‍🏫 Widely applicable in various AI domains, including NLP, speech recognition, visual recognition, and recommendation systems, model distillation is a training technique that trains small models to be as accurate as larger models by transferring knowledge. 🔤 A classic example of model compression can be seen in various BERT models that employ knowledge distillation to compress their large deep models into lightweight versions of BERT. For instance, BERT-PKD, TinyBERT, DistilBERT, and BERT-BiLSTM-based model compression techniques solve multilingual tasks using lightweight language models. – What’s the #11DaysofInferenceAccelerationTechniques? The Deci team is posting, for 11 days, a series of inference acceleration techniques for deep learning applications. If you’re looking for practical tips and best practices for improving inference, follow Deci AI so you won’t miss an update. #deeplearning #machinelearning #neuralnetworks #computervision

14

#RAG — Retrieve Augmented Generation, is a process to optimize the output of large language models by referencing authoritative knowledge repositories beyond their training data sources before generating responses. Large Language Models (LLMs) are trained on vast amounts of data and employ billions of parameters to generate raw output for tasks such as answering questions, language translation, and completing sentences. RAG extends the already powerful capabilities of LLMs to specific domains or internal knowledge bases of organizations without the need to retrain the model. This is a cost-effective approach that can improve the output of LLMs, keeping it relevant, accurate, and useful across various contexts. Why is RAG important? LLMs are crucial AI technologies underpinning intelligent chatbots and other Natural Language Processing (NLP) applications. Their aim is to create bots capable of answering user queries across various contexts by cross-referencing authoritative knowledge sources. However, the nature of LLMs technology leads to responses that can be unpredictable. https://lnkd.in/eD4ZjNj3

1

Polars recently made a few updates to its latest TPC-H benchmark results. After benchmarking several libraries like Polars, Pandas, PySpark, DuckDB, Dask, and Modin, it was found that Polars is the best raw performer. This performance can be compared to DuckDB's in terms of speed. Want to learn more about the TPC-H benchmark update? Read more here: https://lnkd.in/eHfWKwAz #Polars #Python

1

𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹 𝗰𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁 𝗼𝗳 𝗮𝗻𝘆 𝗔𝗜? AI systems won't work at all, if not for the crucial aspect that is data annotation. Data Science UA delivers dozens of ML- and CV-based projects, and we can say for sure - the role of data processing team is hard to overestimate. 𝗪𝗵𝘆? 👉 Approximately, 80% of professionals' time is spent on data preparation. 👉 Properly labeled data ensures the accuracy of safety solutions across the industries. 👉 Data annotation is the backbone of unbiased and fair AI models. 👉 Errors in data annotation can increase project cost by 50%, due to the need for rework. 👉 Quality datasets enhance AI systems' performance by up to 20%. Data Science UA team always strives for the highest data quality standards, so that businesses gain the best results from our AI solutions 🦾 -- #AI #BigData #DataAnnotation #DataScience #MachineLearning #ComputerVision #Technology #Innovation #Tech #ResponsibleAI #DataforBusiness #AIforBusiness #DataScienceUA

5

1 коментар