Amnon Bleich

I call it a "three-part inference plot". When you want plot results of inference (both point + interval estimates, result of testing hypotheses), consider following the approach that has been helping me for long years. You don't have to combine (stack) all the plots together (like in my example) as it may result in a barely readable result (too cluttered, messy). You can display just 3 separate plots: 1) descriptive statistics: mean + SD, or median + IQR (OR) or whatever works. Show what's going on with the raw data. 2) individual point estimates; naive or model-based. But DO NOT use it for deciding if some comparisons yield statistically significant results just by observing if the CIs overlap! This is incorrect and will mislead you often. 👨‍🏫 If you asked now "why?! I've been doing it since I remember", it's a perfect moment to Google a bit and revise this topic :) 3) the results of comparisons itself: CI for the difference (or ratio), p-values, notes about adjustments (if used). This provides a complete picture of the situation and improves transparency. Tweak it as you like, choose the colors you like, strip out unnecessary elements (Tufte's way), but anyway provide the 3 pieces of information. 💡 PS: Usually I replace the upper plot with the ➡ rain-cloud plot ⬅ and overlay it with the 👉 number of observations (very important), but in this very case there were separate plots and tables with summaries, so the Sponsor asked me to skip it. But nothing prevents you from adding some basic numbers, like N, mean (if of special importance), skewness (sometimes it's very important to focus on the shape of the empirical distribution) and whatever else you like. Just don't make it too messy! #statistics #datascience #research #abtesting #abtest #biostatistics

89

12 Kommentare

Causal inference packages in R. Next week, i will be sharing a List of top Causal inference packages in R. This time focusing on Statistical Causal inference packages for Real World evidence / Observational data. Topics covered : Instrumental variables Propensity scores Inverse probability weighting Matching Mediation Mendelian Randomization Genomics Where ? 'Advanced Statistics and Data Science' - Linkedin Newsletter More in depth tips and perspectives adatascience.com #stats #r #rstats #biostatistician #datascience #clincialresearch #realworldevidence #rwe #genetics #genomics #causalinference #causality

68

2 Kommentare

Lately I've been exploring causal survival analysis. The simsurv R package has been a gamechanger. Why? It simplifies simulating survival data. This means I can test out different distributions and complex covariate relationships. I can also explore different hazard functions and time-dependent effects. The vignettes are super helpful as well! I've been able to deepen my understanding of time-to-event data using simsurv. Highly recommend if you're working with time-to-event data! Or if you're just curious about methods for survival data. #CausallyCurious #RealWorldEvidence #CausalInference #RealWorldData [Link: https://lnkd.in/e8xfuBuN]

87

2 Kommentare

CellAgent is a LLM-driven multi-agent framework that simplifies the challenging task of scRNA-seq data analysis. By incorporating roles such as Planner, Executor, and Evaluator, CellAgent effectively decomposes tasks, executes them sequentially, and iteratively optimizes solutions to ensure high-quality results. It shows impressive performance in batch correction, cell type annotation, and trajectory inference compared to existing tools. The framework employs AI techniques, including hierarchical decision-making and self-iterative optimization, to automate data analysis without human intervention. CellAgent represents a significant step towards automated biological data analysis. #genai #agents #llm #singlecell

5

2 Kommentare

Useful new R package for those interested in traditional Cox regression and ML survival models.

3

Did you know you can use "and" and "or" in #cpp to make your code more readable just like Python? It's a really simple language compared to say Forth, Prolog, Rust or Robol. It's very beginner friendly once you understand the basic concepts of pointers, pointer dereferencing, smart pointers, pointer-to-member operations, captureless lambdas, move semantics, semaphores, condition variables, stack and heap allocation, RAII, type erasure, static polymorphism, virtual constructors, and some other easy to grasp basic language features.

104

47 Kommentare

Next is Moritz Schaefer on Multimodal learning of transcriptomes and text enables interactive RNA-seq data exploration with natural-language chats https://lnkd.in/ePBwUj_h #biodata24

2

I'm in Vienna this week attending #ICML2024 and also presenting at the #AI4Mat workshop on July 27th! If you are around, check our poster under the abstract track: Leveraging Fragment-Based Representations in SAFEBert: A Transformer Model for Enhanced Molecular Modeling Molecular language models are demonstrating increasing importance in facilitating rapid prediction of molecular properties, which is essential for innovation in drug discovery and material design. This study introduces SAFEBert, an innovative transformer encoder model that integrates the Sequence Attachment-based Fragment Embedding (SAFE) notation to enhance molecular modeling. SAFEBert incorporates three distinct attention mechanisms tailored to exploit the fragment-based nature of SAFE: intra-fragment attention to capture local relationships within fragments, inter-fragment attention to model interactions between fragments, and disconnected fragment attention to account for non-adjacent fragments. A specialized CLS attention mechanism aggregates the outputs of these attention heads, providing a unified representation for downstream applications. This alignment of the model's architecture with SAFE's inherent fragment structure introduces an inductive bias towards chemically meaningful representations, which we hypothesize will substantially improve performance on various molecular prediction tasks compared to existing models. Furthermore, the attention maps generated by SAFEBert offer insights into the molecular interactions at the fragment level, similar to the way molecular fingerprints are used, potentially enhancing the interpretability and application of molecular language models in real-world scenarios. Nelson Fernandez-Pinto, Entao Yang, Zachary Wilson, Lucy Chen, Sihong Zhao, PMP

38

2 Kommentare

Is GPT-4o good at table recognition? Not quite! OCR looks fine, but tripped by spanning cell table structure! 🤩 However, our UniTable, with only 124M param trained by self-supervision, shows great understanding! 👉 UniTable paper: arxiv.org/abs/2403.04822 👉 UniTable code: https://lnkd.in/e4a7YMxJ #MachineLearning #GPT-4o #Table #Document

85

8 Kommentare

https://lnkd.in/ecDMJvF4 Seqmon: A Lightweight Sequencing Monitor Seqmon is a web application built on Plotly Dash designed to monitor sequencing runs across multiple sequencers and analyze the statistics of demultiplexing processes. Recommend anyone who manages lots of NGS data to try it.

2

Hi folks, Here's a postdoc position some of you may find interesting - check it out! It's a 2-year postdoc position at iDiv (Germany). The application deadline is 13 June (23:59 CEST). The position ‘Open-science tools for monitoring ecosystem extents’ will contribute to the ESA-Funded project "World Ecosystem Extent Dynamics" (WEED), in which the consortium will collaborate with monitoring agencies on four continents to develop fully cloud-based, open-science toolsets that support countries in developing ecosystem maps and time-series, for use in SEEA ecosystem accounting, GBF ecosystem indicators, and other applications. The focus of this position is on developing a data cube of multiple ecosystem characteristics, validation pipelines, and uncertainties. They are searching for candidates with relevation programming/modelling experience (e.g., scalable geo-computation, spatial/spatiotemporal predictive modelling using Machine Learning/AI tools, analysis of land-surface changes) and, ideally, with a good understanding of ecosystem-related concepts and major characteristics of different ecosystem types. Details on the position and application procedure are provided here: https://t.ly/H3NxU

24

When you refer to the 𝗖𝗲𝗻𝘁𝗿𝗮𝗹 𝗟𝗶𝗺𝗶𝘁 𝗧𝗵𝗲𝗼𝗿𝗲𝗺 (𝗖𝗟𝗧), remember - there are MORE than one version of it. Always specify which one you mean: ➡️ Lindeberg-Levy the oldest one; needs IID variables and finite mean and variance. In other words - it's quite limited (independently and same distributed random variables) but has simple assumptions offering faster convergence in distribution - no free lunch 🤷‍♂️ ➡️ Lyapunov relaxes the need for same distributions, but has strong assumption: finite first 2 moments and sufficiently fast vanishing higher moments as N grows (meaning "not to heavy tails" ) - no free lunch 🤷‍♂️ ➡️ Lindeberg (-Feller) as Lyapunov, relaxes the need for identically distributed variables, but has a stronger condition, that no individual variable dominates the sum. Or differently - that contributions to the total variance from variables showing large deviations from their means variable vanish as N grows. BTW, if Lindeberg condition holds, then Lyapunov one holds too. ➡️ Gibrat's law aka Multiplicative CLT for IID log-normally distributed variables. Naturally uses the product rather than sums, leading to the log-normal distribution. Or the normal one for log(sum). ➡️ Generalized Kolmogorov-Gniedenko LT about converging to alpha-stable distributions (Gaussian is a special case, but not the main target here!) for IID variables but do NOT require finite variance and allows for heavy-tailed ones (Levy stable ones). at least as far as I recall... ➡️ CLTs mentioned in the Wither's and Orey's papers - allowing for dependent variables. (e.g. read https://lnkd.in/dBHdEbJU ) Also, check out this interesting article about thinking about CLT as convolutions: https://lnkd.in/di44fQpe Plus 7 cool facts about the CLTs!: https://lnkd.in/ddGDsSn9

62

4 Kommentare

Use LangGraph to re-implement my DrugBot for the multi-model DuckDB. It organizes the LLM workflow as a graph. Now the chatbot can recall past conversations, user-defined concepts, mimic successful interactions and more. https://lnkd.in/gN7Z2Thu #LLM #LangGraph #LangChain Hong Wang Ollie Zaring Eduardo Cerritos Shawn Shih-yun Chen Johnny Lin Shubhanshu Yadav

14

1 Kommentar

🚀 Advancing Information Retrieval with APEER: Automating Prompt Engineering for LLMs! 🚀 I recently delved into a fascinating paper on APEER (Automatic Prompt Engineering Enhances LLM Reranking), a method transforming the way we approach relevance ranking in Information Retrieval (IR) using Large Language Models (LLMs). 🌟 🔍 Challenges in IR with LLMs: Current IR methods with LLMs often rely heavily on human-crafted prompts for zero-shot relevance ranking. This process is time-consuming, subjective, and lacks scalability. The complexities of integrating query and passage pairs and conducting comprehensive relevance assessments further complicate the effectiveness of existing methods. 💡 APEER: A Game-Changer APEER addresses these challenges by automating prompt engineering through iterative feedback and preference optimization. Here's how it stands out: Feedback Optimization: Continuously refine prompts based on performance metrics. Preference Optimization: Improves prompts by learning from positive and negative examples. 🌟 Key Highlights: Reduced Human Effort: APEER minimizes the need for manual prompt crafting, making the process more efficient and less reliant on human expertise. Enhanced Performance: Demonstrates significant improvements in LLM performance on IR tasks, with notable gains in metrics like nDCG@10. For example, APEER achieved an average improvement of 5.29 nDCG@10 on eight BEIR datasets over manual prompts on the LLaMA3 model. Better Transferability: Shows consistent outperformance across diverse datasets and LLM architectures, including GPT-4, LLaMA3, and Qwen2. 📊 Robust Validation: APEER has been rigorously tested on multiple datasets such as MS MARCO, TREC-DL, and BEIR, ensuring its robustness and effectiveness in various IR scenarios. 🌐 Implications: This advancement represents a significant step forward in optimizing LLM prompts for complex relevance ranking tasks. By reducing manual intervention and enhancing the efficiency of LLMs, APEER paves the way for more scalable and accurate applications in real-world IR tasks. 🛠️ Conclusion: APEER's automated approach to prompt engineering is a major leap towards more effective and scalable IR systems. It highlights the potential of integrating iterative feedback and preference optimization to refine LLM performance, offering promising avenues for future research and practical applications. Looking forward to seeing how APEER influences the landscape of Information Retrieval and beyond! #InformationRetrieval #LLM #AI #MachineLearning #NLP #DataScience #Research

1

Shashank Sonkar and colleagues are back with another great paper and matching GitHub repo, "Pedagogical Alignment of Large Language Models". You can read the paper on arXiv here - https://lnkd.in/enZfCJgi, and browse the code here - https://lnkd.in/gyW5hSzp. They demonstrate an interesting approach to generating synthetic data to train open weights models to behave more pedagogically, as well as developing some new evaluation metrics. I think there's lots here for those of us interested in this kind of work to learn from and build on. (If synthetic data generation is new do you, check out this post from answer.ai earlier this week - https://lnkd.in/ecsG7R4b.)

38

4 Kommentare

My first real post on substack. How to solve an annoying error on Windows when using CTranslate2 and pytorch at the same time. Useful for me to remember in the future and hopefully useful for some of you! #substack #softwaredevelopment #deeplearning #dependencies

2

What is a good way to represent a genome which expands itself into a list of arbitrary length seq2seq transformation instructions. So given a vector of any length, the genome should produce transformation instructions to mutate that vector into another (different sized) vector. I want to use a Evolutionary Algorithm to select genomes that produce these transformation instructions. The transformations can be anything but follow "predictable"(in the human sense) patterns. The hope is the final most fit organism genomes are few-shot learners of any arbitrary seq2seq transformation pattern. Currently I am thinking of the genome as a "paramterization" of a very broad transformation function which utilizes parameters for deterministic mutation.

1

1 Kommentar

🥰 Thank you, Lucy McCarren, for allowing me to support the DIS exchange program and guide your Computer Science and Data Science students in exploring the complex challenges of AI ethics. Thanks as well to the DFKI ethics team and the Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) in Berlin for hosting and enabling me to lead the workshop "Beyond Data: Understanding its Meaning." Together, we delved into critical topics in data ethics, such as: 🔹 The impact of cherry-picking data and the responsibility it entails. 🔹 Going beyond correlations to uncover causal relationships. 🔹 Navigating algorithmic black boxes and considering the role of human judgment. 🔹 Quantifying emotions as data – can and should it be done? 🔹 Understanding human behavior through data – what are the limitations? 💡 These discussions are essential in shaping data scientists who are not only technically skilled but also ethically aware. #AIEthics #DataEthics #ResponsibleAI #DataScience #ComputerScience #CausalInference #ResponsibleData

49

2 Kommentare