Original Source Here I have recently joined Kaggle competition, Bristol-Myers Squibb — Molecular Translation (BMS competition) . Unfortunately, I missed solo gold, but I could get some interesting findings, which are also generally useful for common OCR tasks. I would like to share them on this post by taking up Manga OCR as a subject. About BMS competition In BMS competition, participants predict InChI text, which is uniquely defined for each molecule, from printed molecule image. Predict InChI from Image — BMS competition Some people call it as image captioning. Others call it as OCR. Anyway, we predict a sequence from an image. Top solutions are listed here . The most commonly used architecture is a typical transformer encoder-decoder model, which is a kind of combination of Vision Transformer and BART . It seems that Swin Transformer has shown good performance as an encoder. Vision Transformer + Autoregressive Decoder General OCR and Deep Learning The typical e

Original Source Here 假如可以直接 train 一個 classifier，為什麼我們需要 metric learning？ 在 AMMAI 課堂上，我學了多種 metric learning 的方法，metric learning 的目的是希望學一個好的 embedding, 可以有效的把相同 class 的 sample 聚在一起，不同 class 的 sample 盡量分開。但是假如我們已知有 N 種 class ，那 metric learning 跟直接 train 一個可以 predict N class 的 classifier 差別在哪呢？他的優勢是什麼？ 我發現在 stack overflow 上 2017 年就有人討論過這個問題： Why do we use metric learning when we can classify 整理一下底下的留言大概有幾個情境是 metric learning 可以做得更好的： unbalanced dataset: 假如某些 class 的 sample 數量特別少，那 training classifier 可能會完全忽略這些 class。但是 metric learning 因為是針對 sample 計算距離，比較不會受影響。 one-shot/few-shot learning：當出現一個 新的 class 只擁有 一個/少量 sample 時，metric learning 可以更輕易的去 adapt，學習新的 class，但是一般的 classifier 沒有辦法輕易的 extend 到新的 class。 metric learning 是 generative model, 而 classifier 是 discriminative model。 或許可以說，metric learning 只是因為他可以把 class 分開，所以可以“順便”拿來做 classification 的問題，但他還可以有其他的應用。他的功能是更強大的，因為可以把 input space 轉換到另一個 feature space, 使得 classifier 更容易工作。 因為這些特性，所以 metric learning 很常被應用在人臉辨識上。假如是使用一般的

https://miro.medium.com/max/1200/0*FhgV2fS2zEKcm5S6 Original Source Here Make-A-Monet: Image Style Transfer With Cycle GANs By: Colin Curtis , Adhvaith Vijay Try out the web application for the project here: https://make-a-monet.herokuapp.com/ Introduction Here at Research@DataRes we always try to push the limits of deep learning and keep up to date with all the developments in the field. While there is no doubt that the advancements in the academic side of recent ML research is very impressive, there is little real world application for these models unless they can be deployed in a production environment. This is why for our latest project we decided to make sure we can integrate our final model with a deployable application that anyone can use. Comput e r Vision (CV) has without a doubt experienced a renaissance during the past decade, and perhaps the Generative Adversarial Network (GAN) is one of the most captivating examples of modern deep learning. The idea is quite

Original Source Here By AI Trends Staff   AI is having an impact in business, government and healthcare. But nowhere is it having more impact than for the biggest companies with the most resources.   Advantages big companies have include access to lots of data and funds to buy smaller companies with the expertise to do something innovative and profitable with the data. Each company has had to decide on the best way to leverage AI for their business.      Ed McLaughlin, Chief Emerging Payments Officer, Mastercard “The question is how do you use AI right or use it wisely,” stated Ed McLaughlin, Chief Emerging Payments Officer for Mastercard , at the recent EmTech Digital event on AI and big data, as reported in MIT Sloan Review . “The biggest lesson learned is how to take these powerful tools and start backwards from the problem,” McLaughlin stated. “What are the things you’re trying to solve for, and how can you apply these new tools and techniques to solve it better?”   M

Original Source Here By Lance Eliot, the AI Trends Insider   Have you ever glanced at a snapshot and asked someone when they took that photo? I’m sure that you have. You wanted to place the picture into a context of date and time. Maybe the photo was snapped years ago and showcases the past. Or perhaps the picture is quite recent and displays the way things are today. All in all, knowing when a photo was taken can be useful and at times essential.       In the computer field, we often refer to timestamping things.       When a computer is hooked up to a camera, the taking of a picture is usually accompanied by adding a timestamp to the collected image. The timestamp merely indicates the date and time of the picture. This can be stuffed inside the data that contains the actual image or might be added as a supplemental piece of metadata that otherwise describes or indexes the photo.   If a series of photos are being taken, the timestamp begins to be extremely important.   Imagin

https://venturebeat.com/wp-content/uploads/2014/01/cloud-lock-security-wavebreakmedia-shutterstock.jpg?w=1200&strip=all Original Source Here Where does your enterprise stand on the AI adoption curve? Take our AI survey to find out. Micro Focus subsidiary CyberRes this week unveiled the new Amazon Macie integration of its Voltage SecureData data security solution. The new capabilities allow Voltage SecureData customers using AWS to automate a host of cloud security processes for data lifecycle management, including compliance with relevant data privacy and security regulations and standards like PCI DSS, HIPAA, and GDPR. Amazon Macie, launched in 2017 , is the artificial intelligence (AI)-based data security and data privacy service layered into AWS that leverages machine learning and pattern matching to automate laborious processes like sensitive data discovery at scale in the cloud. AWS customers can search, filter, and take actions steps based on Macie alerts through

https://venturebeat.com/wp-content/uploads/2018/02/shutterstock_777026485-e1625048155126.jpg?w=1200&strip=all Original Source Here Where does your enterprise stand on the AI adoption curve? Take our AI survey to find out. Artificial intelligence is transforming how people work by boosting efficiency and productivity. Human resources departments are using AI to create a more adaptive, flexible, and fluid workplace, one where staffers can develop training, streamline onboarding, identify and evaluate candidates during recruiting, process feedback, respond efficiently to service requests, and manage projects. HR is notoriously manual; information is often kept in silos and answering questions can be a labor-intensive process. Whether it is creating workforce experiences personalized to each employee, or sifting through large amounts of information looking for valuable intelligence, HR professionals benefit by incorporating AI into their processes. Jeff Gregory, chief inform

https://venturebeat.com/wp-content/uploads/2021/06/GettyImages-1209420560.jpg?w=1200&strip=all Original Source Here Transform 2021 taking place July 12 – July 15 is not to be missed, but one of the big highlights comes early on Day 2. The Breakfast Series Part 2 , presented by Accenture, is all about decoding the data. Project Lead at Google, Valerie Nygaard, will talk with panelists Anjali Dewan from American Express, Mark Clare of Evernorth, a subsidiary of Cigna Corporation, Arnab Chakraborty from Accenture, and Opendoor’s Ian Wong. Good data is the foundation of AI and machine learning beginning with extracting high-quality data in structured, semi-structured, and unstructured forms. Most critical, however, is ensuring data is relevant to any identified business goals. If not, your ML algorithms simply won’t produce the learnings needed to bring value to your business. And, of course, the risk of bias in data has deservedly garnered much attention – bias can go unnotic

https://miro.medium.com/max/1200/0*l-NijxfMk3lOS_1U Original Source Here 8 Reasons Why Your Topic Modelling Algorithm Performs Worse on Short Text Challenges of topic modeling on microblogs Photo by nadi borodina on Unsplash Short-form text is typically user-generated, defined by lack of structure, presence of noise, and lack of context, causing difficulty for machine learning modeling. Topic modeling aims to identify patterns in a text corpus and extract the main… AI/ML Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot via WordPress https://ramseyelbasheer.io/2021/06/30/8-reasons-why-your-topic-modelling-algorithm-performs-worse-on-short-text/

Original Source Here Open engineering consortium MLCommons has released its latest MLPerf Training community benchmark results. MLPerf Training is a full system benchmark that tests machine learning models, software, and hardware. The results are split into two divisions: closed and open. Closed submissions are better for comparing like-for-like performance as they use the same reference model to ensure a level playing field. Open submissions, meanwhile, allow participants to submit a variety of models. In the image classification benchmark, Google is the winner with its preview tpu-v4-6912 system that uses an incredible 1728 AMD Rome processors and 3456 TPU accelerators. Google’s system completed the benchmark in just 23 seconds. “We showcased the record-setting performance and scalability of our fourth-generation Tensor Processing Units (TPU v4), along with the versatility of our machine learning frameworks and accompanying software stack. Best of all, these capabilities will

Original Source Here GitHub is helping developers to speed up and clean up their code with a new AI-powered tool that it calls Copilot. GitHub Copilot uses an AI system from OpenAI known as OpenAI Codex. The system claims to have a broad knowledge of how people use code and claims to be “significantly more capable than GPT-3” in generating code. By drawing context from the code that a developer is working on, the system is able to suggest entire lines or functions. Meet GitHub Copilot – your AI pair programmer. https://t.co/eWPueAXTFt pic.twitter.com/NPua5K2vFS — GitHub (@github) June 29, 2021 Even veteran coders can benefit from GitHub Copilot by using the system to explore new APIs and discover alternative ways to solve problems without having to scour the web for answers. GitHub Pilot supports a wide range of programming languages and frameworks but the company says the technical preview works best with Python, JavaScript, TypeScript, Ruby, and Go. There are curren

Original Source Here How did you decide to go into data science and — more specifically — into the area of data science you’re currently focused on? My journey into data science has been a pretty interesting one. I am an electrical engineer; post-graduation, I worked in a power utility company in the capital city of India. My daily work revolved around power transformers, grids, and substations. However, as luck would have it, I got a chance to work in a department of electronic meters called Advanced Metering Infrastructure (AMI), where we would sift through the humongous electronic-meter data to identify potential fraud or fault in meters. This was a life-changing moment for me, as I started seeing the immense value that data could bring to any business. I started researching and literally opened a pandora’s box, albeit in a positive way. Back then, I wasn’t aware of a field called data science. I wondered if I could fit in or if a person with my experience could bring value