Neural Network Visualization
netron model visulization
- pip install netron
- view model online: Netron
1 | import netron |
or
1 | import netron |
or
just upload the model .pth
to web.
Maketing Techniques
Marketing Techniques -- from MKT Manager Ms. Jorie
Sharing from Marketing Department Leader
Know your customer
Note that: Our delivery is our service -- not just the product (container leasing is not an exclusive business)
- Find something in common with your customer -- eg. habit, logo, color of company, background of slides, use local language
- Find what they really need (apart from your business) -- like the security need, cross-department bottleneck
- Build connection --> business lunch, casual walk
- Breaking ice and build trust --> bring a coffee and talk what they like (eg. special name)
- Cross Department Collaboration --> start from consent
How to communicate
For MKT, staffs are distributed globally. The communication should be precise and clear.
Focus on 1. what is the general case 2. what is the corner case (be honest if u do not know) --> make sure the audience fully understand
Communicate in standard format. Consider using a template or a well organized system. eg. like date formate.
Build communication log --> who did that and why --> that's how to find the original version and find the man in charge.
Knowledge should be well-documented. --> new recruiter training manual / frequent Q&A
Internal collaboration --> show the pros and cons. and let the leader decide
How the decline --> why we cannot do that / we need internal discussion / we need to report leader
Introduction to ChatGPT
Introduction to ChatGPT
大型/超大型对话模型(Large Language Model, aka LLM)是当下AI领域的热点,近年来,出现了许多对话模型,比如ChatGPT、Bard、文心一言等。本文将以ChatGPT为例,为大家解释什么是LLM以及它能做些什么。
Introduction
首先,我们需要了解什么是语言模型?
曾经为了让计算机理解语言并进行对话,工程师们通常需要拥有语言学等背景知识。他们会尝试从自然语言中分析逻辑,构建一套方法来生成对话。然而,随着机器学习的发展,工程师们采用了“神经网络”(neural network)这样的方法,模拟人脑的工作方式,让AI自己学习语言中的规律和逻辑。
为了实现人类类似的对话模型,工程师们构建了一个超大规模的神经网络(虽然远远不及人脑的规模),然后用大量的文本数据让AI自己学习语言的关联性。
简要流程如下(可跳过此部分):
- 输入大量文本数据,随机去掉一些片段,让AI尝试预测缺失的部分,从而学习文本的内在逻辑。(这个过程被称为预训练,aka pre-training)
- 针对不同的任务,对AI进行专门训练,让它学会回答问题、翻译、写代码等,使AI能够胜任多种任务。使得AI能过适应多种任务。(这个过程被称为微调,aka fine-tuning)
这只是一个简单的理解,实际上ChatGPT的训练和推理方式要远比我提到的更为复杂,涉及许多高级技术。
Diet Plan
Diet Plan
Before Breakfast
Supplements: Probiotics
Hydration: 500 ml water
Breakfast (8:30 AM)
- Baozi: 2 medium-sized
- Eggs: 2 (egg whites)
- Supplements:
- Zinc: 40 mg
- Vitamin C: 100 mg
- Vitamin B complex
- Hydration: 500 ml water
Suggestion: Add a piece of fruit or some vegetables for added fiber and nutrients.
Mid-Morning Snack (10:30 AM)
- Fruit: Apple or Banana (consider varying the fruit choice daily)
- Hydration: 500 ml water
Lunch (11:40 AM)
- Lean Meat: 200 g (chicken, turkey, or fish)
- Rice: 150g cooked
- Vegetables: A side of mixed vegetables (e.g., stir-fried or steamed)
- Supplements:
- Vitamin C: 200 mg
- Vitamin D: 2000 IU
- Hydration: 500 ml water
Pre-Workout (5:00 PM)
- Lean Protein Source: 100 g (e.g., chicken breast)
- Carbohydrate Source: A small serving of complex carbs (e.g., sweet potato, brown rice) - optional based on your energy needs
- Fruit: Apple or Banana
- Supplements: Creatine 3000 mg
- Hydration: 500 ml water
Post-Workout (7:00 PM)
- Whey Protein Shake: 30 g
- Carbohydrate Source: Whole grain bread or a banana
- Hydration: Water as needed
Dinner (8:00 PM)
- Fish: 200 g (e.g., salmon or tilapia)
- Rice: 150 g cooked
- Vegetables: A side of green vegetables (e.g., steamed broccoli, spinach)
- Supplements:
- Vitamin C: 200 mg
- Vitamin A: 5000 IU
- Fish Oil
- CoQ10: 200 mg
- Hydration: 500 ml water
Evening Snack (10:00 PM)
- Nuts: 30 g (almonds, walnuts, or a mix)
- Hydration: Water as needed
3D UNet
Understanding and Implementing 3D UNet for Medical Image Segmentation in PyTorch
Introduction to 3D UNet
3D UNet is a powerful convolutional neural network architecture widely utilized for image segmentation tasks, particularly in medical imaging applications such as MRI and CBCT scans. It has proven to be one of the most effective methods for delineating structures within volumetric data.
This article provides an in-depth introduction to the architecture of 3D UNet and presents a PyTorch implementation along with detailed explanations of each part of the code.
Understanding of Combination
Understanding of combinations (aka. Selections)
combination is to select things from a given set of things without considering sequences. Selecting \(r\) items out of a group of \(n\) items is denoted as \(C_n^r\).
The combination formula is \(C_n^r = \frac{n!}{r!(n-r!)}\) where \(0\leq r\leq n\).
To select 3 things out of 5 things with order, there are \(5\times4\times3=60\) permutations, namely \(\frac{5!}{(5-3!)}\).
Among these 3 selected things, there are \(3\times2\times1=6\) permutations, namely \(3!\).
Thus we can solve combination problems with permutation problems! The number of combination that selecting 3 things out of 5 without order is \(\frac{5!}{3!(5-3!)}\) denoted as \(\frac{n!}{r!(n-r!)}\).
Top K problems
TopK Problem
Find the Top K elements in a given list of size \(n\).
Solutions
- sort the list and extract the top k elements -- inefficient; Time Complexity: \(O(nlogn)\)
- only sort the top k elements: bubble; Time Complexity: \(O(kn)\)
- min heap: maintain a min-heap of size k; Time Complexity: \(O(nlogk)\)
- Quick Selection; Average Time Complexity: \(O(n)\)
A Brief Review of Shortest Path Algorithms
Shortest Path
Given a graph find the shortest path from source vertex to end vertex. \(|V|\) denotes the number of vertices. \(|E|\) denotes the number of edges.
Python Filters
Python Filters
1 | >>> a = {'a':1, 'b':2, 'c':1} |