Zhiying Li

A platform to learn and gain insights as per interest. Look into the GitHub Repositories to gain knowledge on Machine Learning and Artificial Intelligence.

2

Do we need to teach Python and SQL to HEOR/HSR students? About a week ago, Mohin Chanpura, MS ran a poll on which programming language is their go-to language [1]. Now the poll is closed, and I got the honor of presenting the results! But first, thanks so much, Mohin and all the friends who participated in the poll! Let's take a look at the results. As expected, R and SAS/Stata (by StataCorp LLC) are the top programming languages for the Health Economics and Outcomes Research (HEOR) and Health Services Research (HSR) communities. However, two interesting observations: - R is more popular overall, while SAS/Stata is more popular among students. I believe this has to do with their school curriculums. I guess many professors are familiar with SAS/Stata. However, when they come out to the industry, they may need to pick up R as R is free. - To my surprise, there are a good number of people who use Python and SQL, whereas those two languages are almost non-existent for students. Again, I believe this has to do with their schoolwork. I guess these two languages do not have a place in the current curriculum. Personally, I think it's better to learn as many languages as possible. They all have their own strengths and weaknesses. I use R for charting, Python for data handling, and JavaScript for making websites. I read SAS and Stata to collaborate with professors/researchers and try to "read" C and Java if I really need to ;) Should the HEOR/HSR students need to learn Python/SQL? I think so. At least they should be exposed to them and not feel afraid of trying them out! So, here's my ask! If you want to teach Python/SQL to the HEOR/HSR communities, I am really interested in being a resource for you all. I would love to chat! Again, thanks, Mohin, for pulling this important survey! [1] https://lnkd.in/dvpM4Yrz

46

15 Comments

Dr. Hui Liu shows how IBridgePy automatically reconnects to the IBKR TWS or the IB Gateway after a daily restart and shutdown: https://lnkd.in/eWUMyH2i #PythonProgramming #AlgoTrading

3

Let's see what Dr. Yue Zhao, the developer of Python Outlier Detection (PyOD) says about the book: “Anomaly detection is a critical technique to identify rare items in risk modeling, security, and healthcare. Dr. Kuo's book provides hands-on guidance on how to use existing tools, such as PyOD, to leverage this technique in your daily data science work. More importantly, this book reviews more than 10 leading detection algorithms, with both algorithm descriptions and detailed code examples.” ~ Yue Zhao, Principal Developer of PyOD I trust you will find the book helpful. Many readers like the eBook edition as shown in the image. This eBook is in a beautiful format for a pleasing reading experience. We have chosen the Teachable platform for the distribution of the eBook without hefty overhead and the distribution to readers around the world. The credit card transactions are managed confidentially and securely. Sample eBook chapters: free https://lnkd.in/eG8HSChZ eBook on Teachable: $22.50 https://lnkd.in/eeMmwpFz Amazon print book: $65 https://a.co/d/c9Yeqph The print edition adopts glossy cover, color print, and the beautiful Springer font and layout for pleasing reading. Its 7.5 x 9.25 inches portal size goes with most of your books in your bookshelf.

22

2 Comments

Introduction to Daft ( ... vs Polars). Distributed DataFrames for Python (built with Rust) https://lnkd.in/g9vTT9zv #daft #polars #dataengineering #dataframes

81

14 Comments

I still remember the early days of my career as a Machine Learning engineer, navigating various sites looking for data -- Github, Kaggle, UCI, you name it. One particular project comes to mind—a predictive analytics tool for early diagnosis of diabetic retinopathy. Despite the importance and potential impact of the project, we hit a major roadblock: insufficient data. The patient data we had was not only limited but also fraught with privacy concerns. This challenge was an eye-opener, underscoring the critical need for synthetic data in healthcare. Now, that I see multiple services are offering synthetic data. Synthetic data is artificially generated rather than obtained by direct measurement. It mirrors the statistical properties of real-world data, making it a powerful tool in scenarios where data is scarce, sensitive, or difficult to collect. Today we have LLMs, especially GPT-4 that can generate domain-specific (textual) data. Many LLMs are even finetuned and benchmarked using synthetic data generated from GPT-4. But, how can you start and what are some of the best practices? Here are some of the key points that you must consider: 1. Define Clear Objectives: Understand the purpose of generating synthetic data. Whether it's for augmenting training datasets, finetuning like I mentioned preserving privacy, or creating data for underrepresented scenarios, clarity in objectives is crucial. 2. Choose the Right Model: Today, there are various LLMs, according to your budget choose them wisely. GPT-4, Claude, Mistral, LLaMA, etc all good but some are expensive and others are free. If you are low on budget and low on requirements then an open-source model with free Kaggle or Colab support will do. But if you are required to generate a huge amount of data then you might need to spend a little on storage. Keep in mind that running open-source LLMs might require you to spend money because they are very large and require extra storage and computing power. 3. Ensure Data Quality: This is where your model will be tested. Synthetic data should mirror the statistical properties of real-world data. Regular validation against real data ensures the synthetic data’s utility and realism. 4. Maintain Privacy: One of the primary benefits of synthetic data is enhanced privacy. Ensure that the generation process and the synthetic data itself do not inadvertently expose sensitive information. In any case research thoroughly and choose the model according to your requirements. I have attached a code snippet that might help you understand how to generate synthetic and also how to design a productive prompt. The link to the original code can be found in the comment section.

19

7 Comments

🎉 Great News from Entropica Labs! 🎉 We are happy to share some of our latest work on #FaultTolerance and #QubitRouting. Congratulations to the entire team - Chiew Shao Hen, Vishal Sharma, Chai Jing Hao, Ezequiel Ignacio Rodríguez Chiacchio and Hui Khoon Ng - for bringing this project to life! 🙌 Our team has demonstrated that, by using reasonably well-behaved Swap gates, one can preserve the fault-tolerance design of an abstract quantum circuit when mapping it to a physical circuit with a different topology. 🚀 This property is particularly relevant when working with the planar surface code on hardware with heavy hexagonal lattices (like the IBM Quantum platforms). We show how our routing schedules preserve the fault-tolerant nature of the surface code error-correction circuit without the need to modify the surface code circuits to have a hexagonal interaction graph. 📄 Find the paper here: https://lnkd.in/gGyexACH ✨ At Entropica Labs we are working hard towards building the software tools necessary to make fault-tolerant quantum programs a reality across all quantum computing platforms. Stay tuned for more results in the coming months as we continue to improve the design of fault-tolerant quantum programs. #QuantumTech #Innovation #TechContent #Milestone #TeamWork #QuantumComputing

63

Golden Retrieval - MM-Embed: Universal Multimodal Retrieval with Multimodal LLMs (#NVIDIA, November 2024) Paper: https://lnkd.in/gpmQBsUe Abstract: "State-of-the-art retrieval models typically address a straightforward search scenario, where retrieval tasks are fixed (e.g., finding a passage to answer a specific question) and only a single #modality is supported for both queries and retrieved results. This paper introduces techniques for advancing information retrieval with multimodal large language models (MLLMs), enabling a broader search scenario, termed universal multimodal retrieval, where multiple modalities and diverse retrieval tasks are accommodated. To this end, we first study fine-tuning an MLLM as a bi-encoder retriever on 10 datasets with 16 retrieval tasks. Our empirical results show that the fine-tuned MLLM retriever is capable of understanding challenging queries, composed of both text and image, but underperforms a smaller CLIP retriever in cross-modal retrieval tasks due to modality bias from #MLLMs. To address the issue, we propose modality-aware hard negative mining to mitigate the modality bias exhibited by MLLM retrievers. Second, we propose to continually fine-tune the universal multimodal retriever to enhance its text retrieval capability while maintaining multimodal retrieval capability. As a result, our model, MM-Embed, achieves state-of-the-art performance on the multimodal retrieval benchmark M-BEIR, which spans multiple domains and tasks, while also surpassing the state-of-the-art text retrieval model, NV-Embed-v1, on MTEB retrieval benchmark. Finally, we explore to prompt the off-the-shelf MLLMs as the zero-shot rerankers to refine the ranking of the candidates from the multimodal retriever. We find that through #prompt-and-reranking, MLLMs can further improve multimodal retrieval when the user queries (e.g., text-image composed queries) are more complex and challenging to understand. These findings also pave the way to advance universal multimodal retrieval in the future." #genai #retriever #multimodal #search #information Zachary Huhn Summer Crenshaw Scott Phillips Lohith Srikanth Pentapalli Kelly Cohen

7

3 Comments

Bookmarking this will definitely help you in the future!

1

PoLL: Evaluating LLMs with a Panel Approach The challenge of evaluating Large Language Models (LLMs) has grown alongside their increasing sophistication. Traditional metrics often fall short in capturing the nuances of free-form generation, and relying on a single large LLM as a judge, while popular, presents issues of cost, bias, and potentially unnecessary scale. Panel of LLm evaluators (PoLL): This new research proposes a novel approach to LLM evaluation, utilizing a panel of smaller, diverse models instead of a single large judge. The study explores this method across various tasks and datasets, revealing several key findings: * Enhanced Accuracy: PoLL demonstrates a stronger correlation with human judgments compared to single large LLM judges, like GPT-4. * Cost-Effectiveness: Implementing PoLL proves to be significantly more cost-efficient, exceeding seven times the cost savings compared to using a single large judge. * Reduced Bias: By incorporating a diverse set of models from different families, PoLL effectively mitigates the issue of intra-model bias often observed in single large judge evaluations. * Questioning the Need for Large Judges: The research suggests that relying on the largest, most powerful LLMs for evaluation may not always be necessary, as PoLL demonstrates comparable or even superior performance. This work opens exciting new avenues for LLM evaluation, offering a more accurate, cost-effective, and unbiased approach. By leveraging the strengths of a diverse panel of smaller models, PoLL effectively addresses the limitations of current methods and paves the way for a more reliable assessment of LLM capabilities. Read more about the "Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Models" in this research paper: https://lnkd.in/gbJwwi-s #LanguageModels #LLM #AI #NLP

24

𝗨𝗻𝗹𝗼𝗰𝗸𝗶𝗻𝗴 𝗔𝗜 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: 𝗛𝗼𝘄 𝗠𝗶𝗻𝗶𝘁𝗿𝗼𝗻 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝘀 𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 𝗶𝗻𝘁𝗼 𝗟𝗲𝗮𝗻 𝗣𝗼𝘄𝗲𝗿𝗵𝗼𝘂𝘀𝗲𝘀 (Paper link: https://lnkd.in/eJX8emrQ AI Generated Podcast: https://lnkd.in/eRUd-crt) NVIDIA researchers have developed the Minitron approach to compress large language models (LLMs) like Llama 3.1 8B and Mistral NeMo 12B into smaller, more efficient versions with 4B and 8B parameters, respectively. This is achieved through 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲𝗱 𝗽𝗿𝘂𝗻𝗶𝗻𝗴 and 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗱𝗶𝘀𝘁𝗶𝗹𝗹𝗮𝘁𝗶𝗼𝗻, enabling significant reductions in model size without substantial performance loss. 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: • Pruning Strategies: Two methods were explored: depth pruning and joint hidden/attention/MLP (width) pruning, effectively reducing model complexity. • Teacher Correction Phase: Introducing a fine-tuning step allows the teacher model to adapt to specific data distributions, enhancing the distillation process. • Open-Source Models: The resulting models, including Mistral-NeMo-Minitron-8B, are available on Hugging Face with permissive licenses, promoting accessibility and further research. This advancement offers a practical pathway to deploy powerful language models in resource-constrained environments, broadening their applicability across various industries. #ArtificialIntelligence #MachineLearning #ModelCompression #AIResearch #OpenSourceAI #EfficientAI #LLM #TechInnovation

7

DOCTR-L, or how I learned to stop worrying and love offline RL. DOCTR-L stands for Distributional Offline Continuous-Time Reinforcement Learning. Offline Reinforcement Learning is arguably the most common use case for reinforcement learning in practical applications. While offline RL is more widely encountered than online RL, it is also more difficult than the latter. This is because in offline RL, a learning agent can't explore its environment to optimize its policies, and has to rely exclusively on a fixed dataset collected from actions of some other agent (or agents) who could pursue altogether different goals. Having a fixed dataset makes offline RL similar in certain ways to supervised or unsupervised learning. However, in offline RL we do *not* observe an optimal policy, or alternatively actions corresponding to such policy. This means that offline RL amounts to inference or learning from data, but of a very peculiar type. In a recent paper published in Neural Computing and Applications, I showed that with a continuous-time formulation, risk-sensitive or distributional RL with stochastic policies can be cast into a 'soft HJB equation' - a soft relaxation of the classical Hamilton-Jacobi-Bellman equation. The 'Deep DOCTR-L' algorithm proposed in this paper uses a deep neural network to approximate the optimal solution of the soft HJB equation and learn it from offline data. Instead of relying on value iteration or policy iteration methods, Deep DOCTR-L converts offline high-dimensional data into an optimal policy in one step by reducing it to supervised learning. Furthermore, the distributional RL formulation enables a computable approach to the quality control of obtained policies in terms of both expected returns and uncertainties about their values. The journal version: https://lnkd.in/eYxQMQND ArXiv version: https://lnkd.in/gNxhUQd #distributional #offline #continuous #time #reinforcementlearning

52

3 Comments

Basic LangChain Tutorial: Day 5 of 7 Day 5 of our LangChain tutorial series, the focus shifted to exploring the concept of "Agents." Agents in LangChain harness language models to dynamically determine the sequence of actions to take, making them highly versatile and adaptive. This module covered the core components of agents, including the decision-making process, tools, and the use of language models as reasoning engines. Key Highlights - Introduction to Agents:  - Definition: Agents utilize language models to dynamically decide the next course of action.  - Versatility: Capable of adapting to various tasks and environments. - Core Components:  - Decision-Making Process: How agents determine the actions to take.  - Tools Integration:    - Examples: DuckDuckGo, DataForSeo, Shell (bash).    - Usage: Integrated into agent workflows for enhanced functionality. - AgentExecutor:  - Role: Manages the execution of actions, tool interactions, and task completion. - Practical Example:  - OpenAI's Function Calling:    - Demonstrated the process of creating and executing a simple agent.    - Showcased practical applications and real-world use cases. - Memory in Agents:  - Importance: Retains conversation history for coherent and context-aware interactions. Stay tuned for Day 6, where the focus will shift to understanding LangChain retrieval modules in more detail. Follow along to master this critical technology for GenAI jobs. Day 1 of 7:- https://lnkd.in/gz3BufKS Day 2 of 7:- https://lnkd.in/gJ4827TP Day 3 of 7:- https://lnkd.in/gUfsnk9x Day 4 of 7:- https://lnkd.in/gXGptMVT Post on critical skill gap in GenAI jobs:- https://lnkd.in/gyKZdBxK #LangChain #LLMs #AI #Chatbots #NaturalLanguageProcessing #PromptEngineering #DataScience #MachineLearning #OpenAI #GPT4 #ArtificialIntelligence #NLP #GenAI #Python

8

🚀 Enhancing AI Planning with AgentGen 🚀 Just read an incredible paper on arXiv, "AgentGen: Enhancing Planning Abilities for Large Language Model-based Agents via Environment and Task Generation." This groundbreaking research introduces a framework that uses LLMs to generate diverse environments and planning tasks, significantly improving the planning capabilities of AI agents. AgentGen's innovative approach has led to remarkable performance improvements, even surpassing GPT-3.5 and GPT-4 in some tasks! 📄 Link: https://lnkd.in/g_2eibEC #AI #MachineLearning #LLM #Planning #Innovation #Research #AgentGen #arXiv

8

2 Comments

⭐ Excited to share our latest research paper on manifold data augmentation for tabular/table data titled "TabMDA: Tabular Manifold Data Augmentation for Any Classifier using Transformers with In-context Subsetting". In this paper, we address the challenge of limited tabular data, which often hinders the performance of machine learning models. We introduce TabMDA, a novel method that utilizes a pre-trained in-context model to perform label-invariant transformations and enlarge the training dataset in the encoder's manifold space. Our method is training-free, applicable to any classifier, and improves performance substantially on small datasets. This was a great collaboration with Adrián Bazaga, Nikola Simidjievski, Pietro Lio' and Mateja Jamnik Preprint available: https://lnkd.in/dzdypRjf

63