Weikeng Chen

Risk Aversion and Success at Startups Initially inspired by https://lnkd.in/gSAmq2bs Why did Newton and Einstein fail at stock trading so miserably? https://lnkd.in/gN2kFDWd (Sorry Korean alert) It might be tempting to say they weren't experts in stock trading, but the biggest reason is that their decisions were not driven by their true intelligence. The "fear effect" played a significant role. Here’s what we need to understand: - Handling fear and managing risk are crucial. - You can rely on systematic triggers or become immune to fear. Take Alex Honnold, world star rock climber, for example: - He isn’t affected by fear now. His fear receptor is not reactive. - He's so accustomed to fear that it no longer impacts his flight instinct. - This allows him to think through issues clearly. This could explain why repeat founders are more likely to be successful. Sure, factors like accumulated wealth and community play a role, but I believe the biggest takeaway is that they are 'less' fearful, enabling them to make decisions without being impacted by irrational fear. Reshare if you found this helpful ♻️ P.S. Have you ever overcome a fear that helped you succeed? Share your story below. I'll be making regular posts on Monday this time. Stay tuned!

7

2 Comments

Jensen Huang, the Founder and CEO of NVIDIA, delivered an exciting keynote discussing his latest innovations and upcoming AI trends. We're thrilled that he mentioned GliaCloud Co., Ltd. and InfinitiesSoft Solutions Inc., two startups that SparkLabs Taiwan invested in early on. We believe this is the best time to develop AI, and Taiwan is playing a crucial role. SparkLabs Taiwan is all-in on AI. Let's work together and make a difference. #DreamBigGoGlobal #DBGG _ David Chen Peng Agnes Dewey Hsu Wenyu Chen Mallory (Meng-Yao) Chien Kenny Chen Joe (游詩喬) Yu

271

1 Comment

【The AI waiter who understands me better than my other half💁 Right here at 2024 COMPUTEX】 What’s for lunch? No idea? What’s for dinner? Whatever? 😵‍💫 🤷 Have to guess what your significant other wants to eat every day? 🌏The world’s only 3-in-1 AI Kiosk smart ordering machine will be displayed at 2024 BenQ COMPUTEX. 📱Combining AI robot recommendation, AI multi-language menu, and AI voice ordering three functions. 🧑‍🍳Intelligent recommendation of suitable meals for customers with different characteristics, and quick response to needs. Curious about how the AI ordering machine can solve the dilemma of not knowing what to eat? It’s not a psychological test or a horoscope🔮 Come to the expo to uncover the secrets of AI🧐 #FindBenQatCOMPUTEX Come to the expo and stand up for sustainability♾️COMPUTEX exhibition date: 6/4(Tue.)-6/7(Fri.), TaiNEX 1: L0118(4F) #COMPUTEX #SMART+ #AINOW #BenQ #Qisda #ESG #SustainabilityDevelopmentStrategy #BenQGroup https://lnkd.in/gaud89dH

11

Many thanks to Ting Zou (Tina Tsou) and Tiejun Chen for presenting Shifu as part of the "Open Software on Open Hardware" presentation at ONE Summit by LF Networking Shifu is an open-source Kubernetes-native IoT gateway that virtualizes IoT devices into a set of APIs in the form of Kubernetes Pods. It's the perfect solution to solve IoT interoperability problems at the Edge. You can learn more at: https://lnkd.in/gNRPg7u9 Watch the presentation here: https://lnkd.in/eDy9YQ8p #OneSummit #OpenSource #EdgeComputing #Kubernetes #Shifu

10

Attention all biology students preparing for the HKDSE exam! I have made a fantastic ai chatbot a Telegram that can help you prepare. Check out this Telegram bot for past papers: https://t.me/hkdsebio_bot #HKDSE #Biology #ExamPrep

4

2 Comments

https://lnkd.in/gQFDyWn7 The evolution of AI will definitely accelerate the new drug invention process.

3

https://lnkd.in/gSNbceyd Rui Zhu is a Ph.D. student at The Chinese University of Hong Kong (Shenzhen), co-supervised by Professor Changwen Chen and me. He proposed the "Discriminative Self-Supervised Learning Paradigm," which can significantly accelerate the convergence of the Diffusion Transformer (DiT) model. This paradigm has been implemented and is currently used in the HiDream.ai products. In the era of GenAI, many impressive works come from the industry because, compared to academia, the industry often has richer computing resources, real data, and application scenarios, as well as a more agile work pace. This allows for the rapid validation of ideas under the premise of the Scaling Law and the ability to serve millions of users. Several young members of the Sora team also frequently interned at various companies before graduating with their Ph.D.s, accumulating practical experience. The reasoning behind this is the same. HiDream.ai will present five papers focusing on DiT technology at this year's CVPR, and Rui Zhu's SD-DiT is just one of them. We welcome colleagues interested in "multimodal generative models" and AIGC to join our CTO, Dr. Yao, for discussions during the conference in Seattle. We also welcome students to intern at HiDream.ai and experience the technological breakthroughs of GenAI firsthand!

48

Current Issue: April 2024, Volume 13, Number 2 Identifying Key Terms in Prompts for Relevance Evaluation with GPT Models Jaekeol Choi, Hankuk University of Foreign Studies, South Korea Full Text: https://lnkd.in/gZgvZirP Volume URL: https://lnkd.in/gp4d6HNW Youtube URL: https://lnkd.in/gNw6SJjG #nlp #textmining #datamining #machinetranslation #languageprocessing #machinelearning #sentimentanalysis #artificialintelligence #informationretrieval #bigdata #bioinformatics #datamining #machinelearning #fuzzylogic #textmining #neuralnetworks #engineering #technology #computervision #deeplearning #ontologies #socialmedia #web #analytics #ubiquitouscomputing #naturallanguageprocessing #ai #knowledgerepresentation #multimedia #nlpresearchpapers #nlp2024 #nlparticles #currentissuenlp #gpt #chatgpt

4

#AI #DeepLearning #usbwifi #Ubuntu20.04 在 ubuntu 20.04 找不到 USB Wifi 的問題 wifi dongle not detected in Ubuntu 20.04 https://lnkd.in/gQNi4XDt

2

Y Combinator Winter 2024 Batchの採択企業を紹介第2弾は、「Healthcare」で分類されている、病院向けAIエージェント（2社）、臨床記録×AI（2社）、個別化医療×診断（1社）、神経疾患の診断技術を開発（1社）のスタートアップについてそれぞれまとめました。ぜひnoteをご覧ください。 また、Y Combinator Winter 2024 Batchの採択企業とは別に、脳卒中の早期検知する医療機器を開発しているスタートアップとして、Hike Venturesのポートフォリオ企業の「Zeit Medical」もご紹介しています。ご興味のある方はぜひご連絡ください。 Zeit Medical：睡眠中でも脳卒中の兆候をAIを活用して。検出するスマートヘッドバンドを開発しています。再発リスクもある脳卒中のモニタリングをすることで、早期発見につながります。現在、臨床試験を実施中です Zeit Medical https://lnkd.in/gF_zdUpv ブログ記事はこちら（Here is the blog post）↓ https://lnkd.in/gqQegw2Y   *Although the articles are in Japanese, please feel free to contact me in English if you are interested in the content. I would like to introduce companies from the Y Combinator Winter 2024 Batch. Part 2 summarizes each of the startups categorized under "Healthcare": Two in AI agents for hospitals, Two in clinical records combined with AI, one personalized medicine and diagnostics, and one developing diagnostic technologies for neurological disorders. Apart from the selected companies from the Y Combinator Winter 2024 Batch, I'm also featuring Zeit Medical, a startup in Hike Ventures' portfolio, which is developing medical devices for early detection of strokes. If you are interested in "Zeit Medical", please let me know. Zeit Medical is developing a smart headband that uses AI to detect signs of stroke even during sleep. This device monitors strokes, including those with a risk of recurrence, facilitating early detection. Currently, clinical trials are underway.

24

1 Comment

📣 📣📣FCCM 2024 Program Highlight Series 07!📣 📣📣 𝐌𝐑𝐇-𝐆𝐂𝐍: 𝐀𝐧 𝐄𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐭 𝐆𝐂𝐍 𝐀𝐜𝐜𝐞𝐥𝐞𝐫𝐚𝐭𝐨𝐫 𝐟𝐨𝐫 𝐌𝐮𝐥𝐭𝐢-𝐑𝐞𝐥𝐚𝐭𝐢𝐨𝐧 𝐇𝐞𝐭𝐞𝐫𝐨𝐠𝐞𝐧𝐞𝐨𝐮𝐬 𝐆𝐫𝐚𝐩𝐡 𝐀𝐮𝐭𝐡𝐨𝐫𝐬: Wenlu Peng, Jianjun Chen, Wenjin Huang, Yihua Huang 𝐀𝐟𝐟𝐢𝐥𝐢𝐚𝐭𝐢𝐨𝐧: Sun Yat-sen University 𝐊𝐞𝐲 𝐂𝐨𝐧𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧𝐬: 🌟 Data dimensionality reduction. The proposed MRH-GCN introduced a configurable multiple adjacency matrix fusion strategy designed for HGCNs, aiming to reduce the dimensionality of adjacency matrices. 🌟 Data compression. This paper employed a sparse-matrix-tile space search compression algorithm to maximize the compression of sparse data in the adjacency matrix and feature matrix, reducing data sparsity and improving indexing efficiency. 🌟 Regular array pipeline execution mode. The paper adopted a regular array pipeline designed for the three computation stages of HGCNs, matching memory access and computation time, and reducing on-chip memory pressure. 🌟 Efficient hardware architecture. This paper employed an SpMM systolic array to facilitate the compression algorithm and regular scheduling scheme, achieving massive data parallelism. Also, relation-awareness was introduced in the combination stage to enhance computational efficiency.

14

𝓐 𝓑𝓮𝓪𝓾𝓽𝓲𝓯𝓾𝓵 𝓜𝓲𝓷𝓭 If you were on Chinese social media during the past few days, you surely would not miss the viral topic of a 中专 student, Jiang Ping 姜萍，ranked 12th in the qualifying round of 2024 Alibaba Global Mathematics Competition, alongside students from top institutions such as MIT, Cambridge. Tsinghua, and Peking universities. What is 中专? It means vocational high school and Jiang is only 17 years old and a fashion major trainee at a vocational high school in Jiangsu Province. I took a look at the result of those qualified for the final competition (see comments) and was glad to find two Singaporeans, Dylan Toh (ranked 2nd) and Jeck Lim (ranked 7th), up there but they are PhD students and had participated in International Mathematical Olympiad before, and so they are trained and experienced in such competition. Thus Jiang's achievement is marvellous as it takes a lot of sacrifices to train for such competitions given that everyone is a Maths whiz kid, but Jiang has to do it by herself given that she is not majoring in Maths. Actually some of you might be aware of Gaokao 高考 i.e. "The Nationwide Unified Examination for Admissions to General Universities and Colleges" , but do you know that there is a Zhongkao 中考 i.e. "Junior High School Scholastic Aptitude Examination" which is like a streaming as those with good grades (about half of the cohort) would attend high school to sit for Gaokao eventually, whilst the rest would attend vocational school or become school dropout. Thus Jiang's success stirs up the country as it means that even if you are a 中专 student, you could compete with the best minds in the world. I hope her accomplishment helps to send out a message to the parents that scholastic success is not the "be-all and end-all" for a student, and if one does not succeed in the examinations, it does not mean he/she is hopeless but it is because his/her talent and hard work have not being recognised yet, and eventually one would shine in life. Hopefully this would help to mitigate 内卷 (the phenomenon where individuals or groups in a competitive environment push themselves to extreme levels) in the society. As an aside, below is a question from the qualifying round of 2022 Alibaba Global Mathematics Competition. This is a practical problem and it epitomises how Maths is useful in daily situations. In a summary, the problem is to calculate the probability (expected value) on how many cartons do you need to buy to collect 虎虎生威 before you stop where you choose the cartons randomly. It is definitely very daunting for lay people like us to even understand the answer (provided in the comments) but I hope such question/answer would help to open your mind and heart that Maths is beautiful. 🔗#david_tbk_china ➽ #China posts #Maths #Chinesesociety #Chineseeduction #Chineseculture

23

2 Comments

Last day of COMPUTEX. Will make an effort to attend next year. One thing that I noticed was the lack of mobile devices and robots (I only spotted one). The lack of robots at COMPUTEX and in Taipei in general was jarring once I noticed because Jensen Huang's keynote had a lot of robots, and they are everywhere in Shenzhen. The lack of mobile devices was even more jarring, and I don't know if that was because its a normal COMPUTEX thing The stuff that was at COMPUTEX are things that I should expect in my retail store in a month, but there was nothing that struck me as *you must get this*. The AI PC's were underwhelming once I played with them. They were just laptops with a Microsoft Copilot AI chat bot, and the chat bot was not particularly good for an AI. The booths for AI server vendors were full, and the motherboard booths were also packed. Also at Innovex, I found a few investible startups, and I was impressed by a video about space startups. The trouble is that once I was in the "okay sign me up" there wasn't anyone obvious that I could talk to. The problem is that since everything was server focused, you didn't see anything with cool new form factors. You just had the same motherboard with better specs. Also I think it was data center focused, the laptops and the other consumer items were the same old thing only with Microsoft Copilot installed. I did get into a fascinating conversation with an industral mini-pc maker. When I talked to Intel they were interested in the latest high end chips, but when I talked to the mini-pc guy, he pointed out that the Alder Lake chipset on some low end laptops have Iris Xe that's a perfectly good GPU for AI. Looking online, you see the usual software driver issues, but that gives me something to do if I need to work with my laptop.

4

Congrats Bo-Han Lu on the fantastic presentation of our work at LREC-COLING 2024! A great collaboration with Richard Tzong-Han Tsai. We translated Hokkien-Mandarin/English using LLaMA 2-7B by standardizing the Han writing script and evaluating with GPT-4. Enhancing Hokkien Dual Translation by Exploring and Standardizing of Four Writing Systems https://lnkd.in/gjWPw7fT Bo-Han Lu, Yi-Hsuan Lin, En-Shiun Annie Lee, Richard Tzong-Han Tsai Machine translation focuses mainly on high-resource languages (HRLs), while low-resource languages (LRLs) like Hokkien are relatively under-explored. The study aims to address this gap by developing a dual translation model between Hokkien and both Traditional Mandarin Chinese and English. We employ a pre-trained LLaMA 2-7B model specialized in Traditional Mandarin Chinese to leverage the orthographic similarities between Hokkien Han and Traditional Mandarin Chinese. Our comprehensive experiments involve translation tasks across various writing systems of Hokkien as well as between Hokkien and other HRLs. We find that the use of a limited monolingual corpus still further improves the model's Hokkien capabilities. We then utilize our translation model to standardize all Hokkien writing systems into Hokkien Han, resulting in further performance improvements. Additionally, we introduce an evaluation method incorporating back-translation and GPT-4 to ensure reliable translation quality assessment even for LRLs. The study contributes to narrowing the resource gap for Hokkien and empirically investigates the advantages and limitations of pre-training and fine-tuning based on LLaMA 2.

35

1 Comment

Super proud of Mike and his team. Many people asked me over the last year why long context and reasoning are so important. My answer is simple: long context is, in other words, a much better memory that allows LLM to utilize its ability to comprehend incoming text fully which as we all know is a million miles better than RAG. A very clear use case is with processing huge documentation like corporate wiki. Now, the deal with reasoning over long context is also simple: if models can get an ability to process long context but do that poorly then what's the value is in doing? And that's why a proper benchmark like BABILong is so desperately needed. It helps you to open your eyes on how inefficent the majority of the modern methods for processing long context by LLMs are. It helps you to understand what the real goal is. Let's make LLMs understand long context better together!

18

2 Comments

Today, Sopoong Ventures held the "TYPHOON" Demo Day 2024 in Seoul, South Korea. They announced their investment philosophy and how to grow early-stage startups for 6 months. Seven startups debuted after completing the batch program with Seed-TIPS (Early-stage startups incubating program in collaboration with the government and companies). Choi Kyung-hee, a Partner at Sopoong, presented the keynote about Sopoong's vision. It is the 1st impact investment company in South Korea, established in 2008. They focus on social impact and climate change technology. You can find early access opportunities to make a huge impact lately. Tiny teams make a big bang. #climatechange #CCUS #sopoong #VC 최경희 Kyung-hee (Vesper) Choi https://lnkd.in/g8uJN4jT

8

 A collaborative team from Peking University and Microsoft have proposed xRAG, a system that reinterprets document embeddings—traditionally used solely for retrieval purposes—as features from the retrieval modality, integrating these into the language model representation space. This integration eliminates the need for textual counterparts of document embeddings, achieving a significant compression rate.   The core innovation of xRAG lies in its modality fusion methodology, which allows for the direct incorporation of document embeddings into the language model's representation space without the need for the embeddings' textual content. This approach simplifies the retrieval-augmented generation process and significantly reduces the computational resources required, as evidenced by a reduction in overall FLOPs by a factor of 3.53 compared to uncompressed models. Additionally, xRAG's design ensures that both the retriever and the language model remain unchanged, preserving the plug-and-play nature of retrieval augmentation and allowing for the reuse of offline-constructed document embeddings.   Experimental evaluations across six knowledge-intensive tasks reveal that xRAG achieves an average improvement of over 10%, adaptable to various language model backbones. This performance demonstrates xRAG's effectiveness in context compression and highlights its potential to match the performance of uncompressed models on several datasets. These achievements underscore xRAG's role in pioneering new directions in retrieval-augmented generation, particularly from the perspective of multimodality fusion. Modality fusion and context compression will enable high performance generalized modal models on device and/or linearly scalable retrieval systems.    Arxiv: https://lnkd.in/eWEiy6rS

1