codepor @seanfohn

The bootstrap is a powerful, computer-intensive method for statistical inference. It works by resampling with replacement from a single dataset to create many simulated samples. This allows us to estimate the sampling distribution of a statistic, such as a confidence interval,…

Cross-entropy is a loss function used in machine learning to measure the performance of a classification model. Minimizing cross-entropy makes the model's predicted probability distribution match the true one. It is a cornerstone for training models in various real-life…

In statistics, a sampling distribution is the probability distribution of a statistic obtained from a large number of samples drawn from a specific population. It describes the range of possible values for a sample statistic, like the mean or proportion, and how often each value…

who wants $1000 in GPU credits to build this physics background + experience with latex preferred; collect 100+ quality captioned targets from papers; smell-test captions; captions become inputs; model writes latex; compiled image measured vs target

Andrew Carr 🤸 @andrew_n_carr

a week ago

who wants $1000 in GPU credits to build this physics background + experience with latex preferred; collect 100+ quality captioned targets from papers; smell-test captions; captions become inputs; model writes latex; compiled image measured vs target https://t.co/1cOKlPgPas

2 4 98 85K 37

Machine Learning/Data Science Interview Cheat Sheets by Aqeel Anwar. Check out the comment section for some of the pages attached for reference.

A random walk is a mathematical model in which a path is created by a series of random steps. Each step's direction and length are chosen randomly. It is a stochastic process used to model phenomena where future movements are independent of past ones, a concept known as the…

ศัตรูของ Learning Program / Course ต่างๆ ไม่ใช่ AI แต่คือ “ความรู้สึกไม่เกี่ยวข้อง” แค่ UX และความสะดวกไม่ช่วยถ้าคนเรียนรู้สึกว่าไม่เกี่ยวอะไรกับปัจจุบันและอนาคตของตัวเอง ยุค AI ทำให้คนคาดหวังคำตอบเฉพาะตัวทันที L&D จึงต้องสร้างการเรียนรู้ที่เชื่อมกับชีวิตจริงให้เห็นว่าได้ใช้แน่ๆ

Acts of Kindness ช่วยสุขภาพจิตได้มากกว่าการ Seek Joy ให้ตัวเอง แค่ทำ Kindness เล็กๆ 3 ครั้งต่อสัปดาห์ ก็ลด Depression, Anxiety & Loneliness ได้แล้ว Self-care ทำให้รู้สึกดีได้ แต่ Generosity สร้างสายสัมพันธ์และความสุขที่ยั่งยืน

Adam Grant @AdamMGrant

a week ago

Acts of Kindness ช่วยสุขภาพจิตได้มากกว่าการ Seek Joy ให้ตัวเอง แค่ทำ Kindness เล็กๆ 3 ครั้งต่อสัปดาห์ ก็ลด Depression, Anxiety & Loneliness ได้แล้ว Self-care ทำให้รู้สึกดีได้ แต่ Generosity สร้างสายสัมพันธ์และความสุขที่ยั่งยืน

keyboard_arrow_left Previous keyboard_arrow_right Next

95 1K 5K 361K 2K

Download Image

Self-Search Reinforcement Learning (SSRL) by @Tsinghua_Uni trains LLMs to use their own internal knowledge for search tasks. It shows that LLMs can act as built-in web simulators, reducing reliance on external search engines. So let's dive into this method:

FNet (2022): What if we replace self attention with Fast Fourier Transform? - instead of letting each token ‘vote’ on what matters , mix info across the entire seq using a 2D FFT - cheaper (O(NlogN)), efficient, edge-friendly this was believed to make llms fast and efficient…

keyboard_arrow_left Previous keyboard_arrow_right Next

🎓 MIT's Performance Engineering of Software Systems course is fantastic if you are interested in performance. Topics include perf analysis, algorithmic techniques for high perf, instruction-level & caching optimizations, parallel programming, and more. ocw.mit.edu/courses/6-172-…

everyone says they want to understand LLMs. this repo makes you prove it. you write the attention. you train from scratch. you break it. then fix it. no huggingface. no walkthroughs. #llm

Many applications require the generation of unique IDs on their backend. This task is easy on a single server but harder on a large scale. For example, on a single server, you can use an incremental ID or a function to get the time of the day. However, such strategies fail in…

Hidden Markov Models are statistical models that describe a system with unobserved, or "hidden," states that generate a sequence of observable outputs. The model assumes that the probability of transitioning to a new state depends only on the current state, making it a "Markov"…

การที่รัฐบาลสหรัฐฯ กลับมาใช้ภาษีนำเข้าชิป Nvidia จากไต้หวัน 20% หมายความว่า ต้นทุนของ Nvidia พุ่งขึ้นทันที คำถามคือ Nvidia สามารถผลักต้นทุนนี้ไปที่ลูกค้าได้หรือไม่ เพราะถ้า Nvidia มีอำนาจการตั้งราคาจริง เราจะเห็นบริษัท data center อย่าง Coreweave, Lambda หรือ Hyperbolic…

What is Mixture-of-Recursions (MoR)? It's a next-level version of Recursive Transformer that learns to give each token its own “thinking depth” and optimizes memory use. MoR has a small set of layers it reuses and has 2 main components: ▪️ Routing mechanism: “Decides” how many…

XQuant instead of KV quantization? XQuant by @UCBerkeley rematerializes keys and values on-the-fly with quantized layer input activations. Its advanced version XQuant-CL cuts LLM memory needs up to 12×! Here is how it works:

open source tool for anonymizing and generating synthetic data

Docker Run vs. Attach vs. Exec 🔽 Explore how these key Docker commands work under the hood in my most recent write-up labs.iximiuz.com/tutorials/dock… Why bother: A fundamental understanding of how tech works enables you to apply the knowledge to new domains and synthesize hard skills.

this isn’t another ‘AI hype’ book. it’s a guide for domain experts, engineers, scientists, and problem solvers on how to actually teach machines what you know. > if you want autonomous AI that works in the real world, start here