Get the most recent info and news about AIR on HackerNoon, where 10k+ technologists publish stories for 4M+ monthly readers.

airedefined

It is no longer just buses and trains, traffickers are increasingly taking the air mode to quickly shift victim girls and women from source to destination states.The issue took centre stage at a seminar on the problem of human trafficking organised by National Commission for Women and the Central Industrial Security Force in New Delhi

Need to step up airport vigil as traffickers ‘flying victims’ to destination states’

Delta Air Lines says the pilot accused of threatening to shoot the plane’s captain during a flight no longer works for the airline

Delta says pilot accused of threatening to shoot the captain no longer works for the airline

A first group of civilian evacuees from Gaza crossed into Egypt under a Qatari-mediated deal on Wednesday while Israeli forces bombed the Palestinian enclave from land, sea and air anew as they pressed their offensive against Hamas militants.

First evacuees leave Gaza after another night of Israeli bombardments

- An alphabet book and a notebook lie on top of an elevated wooden floorboard of a middle school in Let Yet Kone village in Tabayin township in the Sagaing region of Myanmar on Sept. 17, 2022, the day after an air strike hit the school.

Researchers hope tracking senior Myanmar army officers can ascertain blame for human rights abuses

The Houthi attacks will be hard to respond to from sea, and the naval version of Iron Dome is designed for self-defense or protecting a nearby ship'

Could Israel’s New Sa’ar 6 Corvettes Give Its Southern Flank Greater Defensive Depth Against Houthi Attacks?

A recent alteration of an NCAA rule could put Harbaugh himself square in the association's crosshairs ... again

Even if unaware, Jim Harbaugh can be held responsible by NCAA for Michigan sign-stealing scandal

On Oct. 31, the inspector general's office for the U.S. Department of Transportation confirmed that plane captain Jonathan J. Dunn has been indicted over an August 2022 incident in which he allegedly threatened to shoot his co-pilot over his decision to divert the plane because one of the passengers on board needed medical help.

A pilot reportedly threatened to shoot his co-captain in the air

It’s the worst nightmare for the people that have to plan an attack and for the people that have to execute it,” says Walker Mills, a Marine Corps captain who has studied and written about tunnel warfare.

The Tunnel War

It has not been an easy coaching transition at Mississippi State, but the Bulldogs have veteran quarterbacks and a tricky defense.

Kentucky football: Five things to know about the Mississippi State Bulldogs

Jonathan Dunn, a former Delta pilot, has been indicted in Utah following an incident in which he allegedly threatened to shoot a captain, officials say.

Former Delta pilot indicted after allegedly threatening to shoot captain during flight

Forward Air (NASDAQ:FWRD) declares $0.24/share quarterly dividend, in line with previous. Forward yield 1.47% Payable Dec. 7; for shareholders of record Nov. 22; ex-div Nov. 21. See FWRD Dividend Scorecard,

Forward Air declares $0.24 dividend

JetBlue Airways Corp (NASDAQ:JBLU) stock plummeted after it reported a third-quarter FY23 operating revenue decline of 8.2% year-on-year to $2.35 billion, missing the consensus estimate of $2.81 billion. Adjusted loss per share was $ (0.39), missing the consensus of $ (0.25).

Air Traffic Control and Weather-Related Delays For JetBlue Was 'Staggering' In Q3, Stock Slides

The Israeli military is carrying out expanded ground operations against Hamas terrorists in Northern Gaza in what Prime Minister Benjamin Netanyahu is now calling the "second war of independence." More than 9,

IDF: Hamas commander who 'directed' October 7 terror attack killed in airstrike

While the snow has stopped falling across Colorado and the sun continues to warm things up, temperatures will stay roughly 10 degrees cooler than average Tuesday.

Colorado weather: A bit of a chill in the air for Halloween trick-or-treating

Russia lost over a dozen precious helicopters to Ukraine's U.S. ATACMS missiles. Now, a tactical and technological race is afoot.

Helipocalypse: Ukraine's New Missiles Blew Up 11% of Russia’s Ka-52 Attack Helicopters

Aerospace pioneer and investor Dómhnal Slattery appointed to Chairman of the SpinLaunch Board of Directors Investor and Bolt General Partner Greg McAdoo joins the SpinLaunch Board of Directors Telecom

SpinLaunch Announces New Leadership Roles to Accelerate the Commercialization of Low-Cost, High-Performance Space Solutions

The second annual special, co-hosted by Sheryl Underwood and Roy Wood, Jr., will be simulcast Nov. 25 on Paramount+ The post TheGrio Awards to Air on CBS, Honoring Mariah Carey, Don Cheadle, Eddie Murphy and More appeared first on TheWrap.

TheGrio Awards to Air on CBS, Honoring Mariah Carey, Don Cheadle, Eddie Murphy and More

Dozens of Palestinians have been killed in an Israeli air strike on a refugee camp in the Gaza Strip, a hospital director has said, as the UN has warned children in the besieged enclave face dying of thirst.

Israel-Hamas war – live: ‘Dozens killed’ in Gaza refugee camp blast as UN warns children face dying of thirst

Wil Gonzalez, director of marketing and Darin Siples, president of GrowGlide, a division of PIPP Horticulture, engaged in a conversation about the company's innovative solutions for indoor cannabis cultivation.

Scaling Profits with Multi-Tier Cannabis Cultivation: GrowGlide's Lego-Like Solution

Although shrouded in official secrecy, it appears the Israeli Defense Forces is close to — if not already — dividing the Gaza Strip in half.

Israeli troops and tanks continue their advance into Gaza

Navigating AI Integration in Retail: The Change Management Compass

AIR is a Canadian startup focusing on a new AI training approach that allows the tackling of challenges that are too complex for traditional machine learning methods. We are doing this by building a toolset for AIs and people to learn continuously from each other. AIR developed Cogment, the world's first open-source framework to provide the means to train people and AI together, in multi-agent, multi-human, multi-rewards contexts. HackerNoon Reporter: Please tell us briefly about your  background. Dorian Kieken (Founder, President): Before founding the AI start-up AIR (formerly Age of Minds), I shipped 17 video games in multiple roles ranging from Design to Production, including the acclaimed games Mass Effect 2 (Game of the year 2010) and Mass Effect 3 (RPG of the year 2012). I co-founded the BioWare Montreal studio and helped it grow to 100+ people. I introduced and evolved agile production methodologies & tools to multiple companies, and drove the implementation of a distributed leadership/teal system. I'm also an active member of the Montreal AI ethics group and a strong believer in taking key actions today to nudge humanity towards a good future. Greg Szriftgiser (Co-founder, Design): I spent 20 years professionally studying, writing about, then finally designing and writing for the video games industry and its exploration of interactivity and dynamic systems. I became particularly interested in procedural storytelling (how to design intelligent systems able to create compelling narratives through various storytelling devices), which led to AI research and a passion for the struggles and achievements of machine learning as a way for humans to achieve more. François Chabot (Co-founder, Technology): I come from a computer engineering background with a specialization in artificial intelligence. Prior to co-founding AIR, I worked in video games (Capcom), mostly building AI and Gameplay systems in heavily resource-constrained scenarios, notably Dead Rising 2, real-time hardware-in-the-loop simulators (Opal-RT), and malware search (Google). I have a great interest in the relationship between people and machines, both when it comes to programmers and people using programs. Nowadays, I see myself as multidisciplinary glue that can bring various disparate systems together. Craig Vachon (CEO): After grad school, I couldn’t find a job, so I started a small company. I got extremely lucky and sold that company to a Japanese firm two years later. After a few more lucky professional experiences (and having lived as an expat for 12 years), I decided to spend my efforts helping fellow entrepreneurs solve important human challenges as an angel investor. I joined AIR because I like to think we’re tackling one of the more impactful issues that humanity faces—how humans will continue to benefit from/steer/trust AI. What's your startup called? And in a sentence or two, what does it do? It's called AIR, which stands for AI Redefined. We create technology for “steerable AI”, leveraging the advent of AI to the benefit of humankind through human-AI collaboration technology. Our open-source AI orchestration platform, Cogment, allows AI practitioners to design, train and deploy complex intelligent ecosystems, mixing multiple humans and artificial agents of various kinds to bring about results that are more context-aware (ethical, empathetic, explainable). What is the origin story? After enjoying great success in the game industry, and what was a dream-job for me with recognition in a fantastic studio, I felt motivation slipping away around 2014. It coïncided with the birth of my first daughter. I discovered that actually, this loss of motivation came from not working on something more meaningful that will help the next generation. Dorian: Among the many challenges this next generation will face (climate change, energy, water, etc.), I identified AI, and in particular, human/AI alignment, as the most important one that I could get involved in. I got strongly engaged in the Montreal AI Ethics group and ultimately, it led to the creation of AIR that is tackling this human/AI alignment problem. I met Dorian around the time I was diving into AI myself, through common friends, and we immediately clicked on several topics, as well as how we thought the AI conundrums could be tackled. Greg: We also shared this desire to see how we could leverage what we had learned in the video game industry to tackle something more impactful with regard to the various global challenges on the horizon. Throughout my career, I had the good fortune to be exposed to a number of fascinating, seemingly disparate domains and technologies. On top of that, trying my hand at the startup process had always been on my radar ever since I left university. Francois: So once I reached a point in my professional development where I was starting to feel confident enough in my skills to take a crack at it, I looked back at everything that had interested me and that I had become good at so far: The way humans connect and interact with machines, Simulations, Machine Learning, etc… and pondered how to bring it all together into something truly novel, interesting, and useful. During that process, I was introduced to Dorian, and the rest is history. I wasn’t a part of the origin of this company. I joined this year (2021) to help productize and accelerate the adoption of the Team’s amazing technology. Craig: What do you love about your team, and why are you the ones to solve this problem? The dual expertise in both the world of AI and the world of simulation/video games (human-AI interaction) makes for a very unique team of people with expertise in more than one area. It's something that makes the team especially well-suited for the problem we tackle because it needs to be tackled from multiple points of view. Dorian: Not just the AI point of view which centers around technology, rewards, methods of training, but also from the environment where agents evolve and interact with each other and with human users. One parallel that comes to mind is the birth of CGI animated movies, which required people with dual skill sets in Art/animation and computer science to truly blossom. Beyond the multiple skill sets people have in the team, there is also a very important cultural aspect in my opinion, of a human-centered perspective on the world. Emotional maturity, going hand in hand with a very high prevalence in each and every member of AIR's team of an almost child-like curiosity in diverse topics makes for a very particular blend of talents and personalities to tackle this long-term vision of synergy between humans and AI. Greg: The team is mission-driven and diverse in terms of its skill sets and expertise. They fully understand and appreciate the opportunity we’ve been given to ensure ‘humans and AI elevate each other.’ Craig: If you weren’t building your startup, what would you be doing? Dorian: Probably finding another way to nudge the future of AI and humans in a good direction. Maybe through another existing company like DeepMind? Or an organization like the Montreal AI Ethics Institute? Dorian: Probably something related to Video Games, Interaction Design, and/or Storytelling, I suppose, but with the missing excitement of working on a truly transformative technology with a mission bigger than myself. Greg: I would probably be doing something very similar: Tackling hard problems that affect how people relate to technology in some way, and doing so in a healthy collaborative environment. The startup is a means to this end for me. I’m not big on entrepreneurship for the sake of entrepreneurship. François: Helping other great founders pursue worthwhile goals. Craig: At the moment, how do you measure success? What are your core metrics? In the long term, on how human-centric the training of AIs becomes, and how successfully we move past the narrow "AI behaving without context" era we are currently in, to reach towards AI with a broader understanding of context. Dorian: In the short term, it's about the scope and type of collaborations our open-source platform enables, and our ability to generate license-based revenue from customers that value context/steerability/trust from their AI results. Craig: What’s most exciting about your traction to date? The collaborations we have with CRL (Chandar Research Lab at Mila) on Hanabi and the University of Alberta on complex multi-agent problems come to mind. Dorian: From a more general perspective, I'm very excited to see more and more thought pieces, actual work, and studies popping up and coming to similar conclusions we did: that by designing intelligent systems with human participation from the get-go, issues like bias, trust, or explainability can be addressed at the same time as the ability for those systems to achieve better results, thanks to the complementarity between the different strengths humans and AI bring to the table. Greg: : One of our aerospace customers admitted we had made him a hero as we accelerated his roadmap in AI by two full years. (How often do you hear that from the head of a Business Unit?) Craig What technologies are you currently most excited about, and most worried about? And why? AI in general. I'm excited about how it can have a positive impact on humanity, and I’m worried about how this impact could also be negative. Dorian: There is no doubt in anyone's mind, I think, that AI can help us solve very important issues, and bring us much-needed help; but used poorly, it can also exacerbate those issues or even create new ones. It's really similar, I think, to any groundbreaking shift of the technological paradigm in which, or with which, we build our societies. Many people will probably talk about the Internet as the most recent example of a profoundly world-changing technological paradigm shift, and they would be right, but even though we now have around 3 decades of hindsight, I think it's fair to say the jury is still out on whether our societies are better for it or not. Greg: AI will probably impact us at least similarly, if not much more profoundly; that is why I think we need to actively work at making it more accessible to everyone, but also more human-centric. I am excited about the future of how we express what we want computer systems to do. Technologies that are being widely deployed right now feel like we are dipping our toes towards expressing intent in a more semantically meaningful manner. Francois: The massive rise of Python as a programming language and Kubernetes as a deployment platform are but baby steps in that general direction. I am convinced the next generation of “telling machines what we want from them” will open the door to some truly amazing things, and that this is coming a lot sooner than most people would expect. As far as worries go, I am particularly concerned about our willingness to deploy socially disruptive systems based on technologies for which we only have proof of concepts exhibiting critical flaws with the assumption that functional implementations are within our grasp. Image-driven emotion detection and blockchain are particularly egregious examples of this in my mind. Today’s narrow AI provides only the thinnest veneer of humanity. And yet, without an iota of context, common sense, empathy, or morality, we want #AI to solve significant human problems. We will do better (with ). Craig: Cogment.AI What drew you to get published on HackerNoon? What do you like most about our platform? Getting nominated as a Startup of the Year was a very nice surprise! HackerNoon offers a very open, accessible platform for people to learn and share knowledge, with accessible tools everyone can use from anywhere; in that sense, it aligns pretty well with how we think about what the future of AI should be :) What advice would you give to the 21-year-old version of yourself? I was already driven by passion back then. Probably something along the lines of "a healthy mind in a healthy body". Something I only learned in my late twenties. Dorian: Better care of the body (training, food, etc…) leads to better sleep which, in return, gives you more time for all the cerebral activities. "Pay more attention to the irrational and the way humans behave, I swear it's fascinating!" Oh, and, I'd also tell myself: Greg: "sleep more, you idiot, stop pulling multiple all-nighters in a row, circadian rhythm is more important than you think". Don’t second-guess yourself so much. The shame of being proven wrong is only fleeting, but the missed opportunities often don’t come back. Francois: Say “yes” more often. The best adventures, (plus laughter/learning) always start by saying, yes. Craig: What is something surprising you've learned this year that your contemporaries would benefit from knowing? From a business standpoint, that the scaling potential of a business and its recurring licensing revenue is more important to investors than the number of revenues! Also, from a cultural perspective, how much face-to-face human contact is important, particularly after the past year of Covid-19 isolation. We are social creatures more than ever. Dorian: The idea in itself is no surprise to me, but the extent to which it verifies surprises me regularly: that what reality is or could be is irrelevant in the face of what human culture dictates it is or should be. I am also surprised by how much more fragmented and inward-facing culture seems to become in the age of the so-called globalization of information. Greg: Learning to skateboard as an adult feels like rebelling against the rules of physics themselves. The sequence of events required for it to barely function is unlikely. Everything about the process is an unstable equilibrium. Yet here you are, spitting in the face of Newton’s third law. Francois: Two things: Hairy, audacious goals are life-affirming; And the sequel (of the comedic spy thriller I published last year) isn’t as easy as it ought to be. Craig: Don’t forget to vote for R as Startup of the Year in AI Montreal.

Tackling a new Human-Centric AI  Paradigm: Startup Interview with AIR Cofounders

AWS AI Services Scratch Extension In my previous , I have mentioned that we should prepare our children for the future. In the future, to succeed our children should have skills like creative thinking and problem solving. STEM education can improve this skills. post According to MIT Professor , coding is the new literacy. A visual programming language called could stimulate the curiosity we know it exists in the children from the day they have born. You can read stories about how it changes children’s lives , and . Mitchel Resnick Scratch here here here As I am working with , I have been wondering how we could teach AI to our children easily. Then I came across , an extension framework to Scratch for adding new functionality to Scratch. With ScratchX, you can write extensions to integrate Scratch with anything you imagine. There are sample extensions for , , and . AWS AI services ScratchX LEGO WeDo Arduino Raspberry Pi Then I thought I can make a Scratch extension for using AWS AI services like Polly and Translate. And it is . here You can try sample project that uses Amazon Polly to speak the text you have entered in the language you choose. Also you can translate text between languages and speak the translated text in target language. here Below is a video recording of the sample project. Developing The Extension If you wonder how the extension is developed, let’s develop the extension step by step together. A minimal Scratch Extension is created like below. After creating the extension object, we should register it by calling ScratchExtensions.register with the name of the extension and a descriptor which defines the blocks that will be created. For more information on making Scratch Extensions, see . here We can develop our extension in three steps: Create the descriptor Implement the methods Try the extension in ScratchX Let’s start. Create the descriptor The descriptor defines the blocks that will be created after the extension is registered. Created blocks will be shown in ‘More Blocks’ section of ScratchX. The blocks of our extension are defined as below. First block is for initializing the AWS JavaScript SDK in the browser. Amazon Polly blocks are the next two blocks. And last blocks are for Amazon Translate service. Parameters of blocks aredefined as ‘say %s’. We can use lists to choose parameters from with ‘%m.sourceLanguages’ syntax, where ‘%m.’ says that we want to use a list and sourceLanguages shows the name of the list. In ScratchX lists are known as menus and menus are defined with ‘menus’ field. Blocks of our extension will look like below. Blocks are shown in ‘More Blocks’ section 2. Implement the methods After we defined our blocks, we will implement the methods that will be executed when blocks are used. For more information on method details, you can look . Usage of AWS services in the browser with AWS JavaScript SDK is explained in detail in . extension JavaScript file my previous blog post 3. Try the extension in ScratchX In order to try our extension, we should load it from a URL. I recommend for this. MiniWeb HTTP Server After you can run the server in your directory at 8080 port, you can try it in your browser by visiting the address below. I recommend Google Chrome browser, ScratchX is not working in Firefox. IE will not load the extension from http: url, although it works over https: url. https://scratchx.org/?url=http://localhost:8080/awsAiServicesScratchExtension.js When ScratchX warned you, click ‘I understand, continue’ to load and try the extension. Summary In this post, I have shown how a Scratch Extension can be created easily for using AWS AI services in Scratch. You can find the code . For more information on making Scratch Extensions, see . here here For more information on how AWS AI services can be used in the browser using AWS JavaScript SDK, please see my previous blog post: You Can’t Fail To Learn A New Language With These AWS AI Services: Translate, Rekognition, Polly Also, you might like my other post about creating a serverless app for saving allergy information with the picture of children and searching the allergy information later with the picture of the children by using face recognition methods of Amazon Rekognition: Serverless Allergy Checker with Amazon Rekognition, Lex, Polly, DynamoDB, S3 and Lambda I will continue to write about different use cases of AWS AI service. I will add object and text detection features to the extension using Amazon Rekognition. I am developing a Chrome Browser Extension for reading the selected text in the page you are viewing using Amazon Polly and detecting Authy authentication codes from the camera using Amazon Rekognition. Please stay tuned or follow. If you like this post, please share, clap or follow. Thanks for reading.

AI Is Hard? It’s Child’s Play With This AWS AI Services Scratch Extension

Google Gemini Did Not Disappoint on These Blockchain x AI Use Cases 

Authors:
(1) VINÍCIUS YU OKUBO, Dept. Electronic Systems Engineering, Polytechnic School, University of São Paulo, Brazil;
(2) KOTARO SHIMIZU, Department of Applied Physics, The University of Tokyo, Tokyo 113-8656, Japan;
(3) B. S. SHIVARAM, Department of Physics, University of Virginia, Charlottesville, Virginia 22904, USA;
(4) HAE YONG KIM, Dept. Electronic Systems Engineering, Polytechnic School, University of São Paulo, Brazil. Authors: Authors: (1) VINÍCIUS YU OKUBO, Dept. Electronic Systems Engineering, Polytechnic School, University of São Paulo, Brazil; (2) KOTARO SHIMIZU, Department of Applied Physics, The University of Tokyo, Tokyo 113-8656, Japan; (3) B. S. SHIVARAM, Department of Physics, University of Virginia, Charlottesville, Virginia 22904, USA; (4) HAE YONG KIM, Dept. Electronic Systems Engineering, Polytechnic School, University of São Paulo, Brazil. Table of Links Abstract and I. Introduction Abstract and I. Introduction II. Related Work II. Related Work III. Methodology III. Methodology IV. Experiments and Results IV. Experiments and Results V. Conclusion and References V. Conclusion and References II. RELATED WORKS A. JUNCTIONS AND TERMINALS DETECTION A. JUNCTIONS AND TERMINALS DETECTION A. JUNCTIONS AND TERMINALS DETECTION Junctions and terminals are shapes with relevance that extend beyond materials. Within computer vision, recognizing and enumerating them has been performed in diverse contexts, such as natural landscapes [20], biology [21] and handwriting images [22]. B. CLASSICAL METHODS FOR JUNCTIONS AND TERMINALS DETECTION B. CLASSICAL METHODS FOR JUNCTIONS AND TERMINALS DETECTION B. CLASSICAL METHODS FOR JUNCTIONS AND TERMINALS DETECTION One prevalent approach for detecting junctions and terminals involves using skeletonization as a pre-processing step. This process reduces the image to one-pixel-wide lines to represent its structures. Points in skeleton can then be identified as terminals, junctions and crossings based on their neighboring pixels. This technique has been applied in vascular images [23] and in the analysis of handwritten Chinese characters [24]. Pre-processing techniques using contour information have also been explored for junction detection. Lee and Wu [25] investigated stroke extraction in Chinese characters. Their method segments regions according to their contour, identifying junctions by counting neighboring regions. Maire et al. [26] proposed a junction detector in natural images by locating intersecting contours. This approach applies an expectation–maximization style algorithm to iteratively select relevant contours and suggest the junction’s position. However, skeletonization and contour finding are noise sensitive processes, and a pre-processing error will lead to a detection error. Junctions and terminals can also be identified by analyzing the arrangement of linear structures within the image. Su et al. [27] describe a technique for identifying these linear structures using the Hessian Matrix. This approach has been validated in biological images such as blood vessels, neutrites and tree branches. Xia et al. [20] present a junction detection method in natural images, based on amplitudes and phases of the normalized gradients of the image. Template-based approaches quantify the similarity of the appearance of image regions and the template. Deriche and Blaszka [28] modeled this approach as energy minimization, which is calculated by the deviation between the image and a predetermined model. This enabled the detection of key image features, such as edges, corners and terminals. C. DEEP LEARNING METHODS FOR JUNCTIONS AND TERMINALS DETECTION C. DEEP LEARNING METHODS FOR JUNCTIONS AND TERMINALS DETECTION C. DEEP LEARNING METHODS FOR JUNCTIONS AND TERMINALS DETECTION Owing to the success of R-CNN based detection techniques, Pratt et al. [21] developed a pipeline for identifying junctions and crossings in retinal vascular structures. Their method involves two main steps: initially, detection regions are proposed centred along the blood vessels, which are then classified as junctions, crossings or background. To generate the detection regions, their approach requires a binary segmented version of the exam. Theses images undego a skeletonization process, with the resulting points serving as references for the centers of the blood vessels. Zhao et al. [29], addressing the same problem, proposed using a Mask R-CNN base model [30] for region proposal. This strategy enables inference without the need for binary segmented version of the exam. However, during training, the segmented images are still used in the Mask R-CNN model to enhance its learning capabilities. This approach surpassed the performance of previous techniques in the detection of junctions and crossings in retinal vascular images. III. METHODOLOGY A. DATASET A. DATASET A. DATASET In this study, we used films of a ferromagnetic material of recognized technological importance, Bi:YIG, and obtained magnetic images of the labyrinthine patterns using a microscope with polarized light [31]. We specifically focused on the evolution of the labyrinthine patterns under the demagnetization field protocol described below. First, we prepared the sample in the fully saturated state by applying a sufficiently large magnetic field in +z direction, which is perpendicular to the films. In this state, magnetic moments in the Bi:YIG film are forced to point in the field direction. Next, we instantaneously dropped the field to zero and hold it for 10 seconds to get the image. Since the magnetic field is zero, magnetic moments can point upward and downward, resulting in the labyrinthine patterns shown in Fig. 1; the bright and dark regions represent the domains with opposite directions of magnetic moments. Such a process involving switching on and off the magnetic fields is considered half of the demagnetization step. In the remaining half step, a magnetic field was applied with a reduced amplitude and oriented in the opposite direction. We again captured the magnetic domain image after reducing the magnetic field to zero. By repeating these protocols up to 18 steps, we investigated the evolution of the labyrinthine patterns in the demagnetization process step by step. The amplitude of the magnetic field was exponentially reduced with each step. We conducted a series of demagnetization processes from the fully saturated state, repeating this cycle six times. Furthermore, we explored another situation, where the magnetic field was initially applied in the −z direction, and its direction was alternated step by step. Consequently, a total of 12 demagnetization processes were performed, yielding a collection of 444 domain images. All measurements reported here were performed at room temperature. The experimentally obtained images covered an area of 2 mm × 1.8 mm. The original high-resolution color images were converted to grayscale and their resolutions were reduced to 1300×972 to facilitate processing. Furthermore, a median filter with kernel size 3 was applied to reduce noise. B. TM-CNN OVERVIEW B. TM-CNN OVERVIEW B. TM-CNN OVERVIEW Our approach to detect junctions and terminals in magnetic labyrinthine patterns consists of two sequential steps: proposal of potential detections and their classification between junction, terminal and false detections. It is inspired by other cascaded object detection techniques like Viola and Jones face detection [8], R-CNN [12], and scale and rotation invariant template matching [32]. Fig. 3 illustrates the overall structure of TM-CNN. In the first phase, the algorithm generates a preliminary set of potential detections. It must propose all true defects, even if it also generates many false positives. We achieve this by applying template matching detection with a low threshold, followed by a non-maximum suppression. In the second phase, in order to eliminate the false positives, each potential detection is filtered by a CNN classifier. C. TEMPLATE MATCHING C. TEMPLATE MATCHING C. TEMPLATE MATCHING The basic form of template matching finds instances of a smaller template T within a larger image I. This is done by calculating some similarity metric between the model T and the content of a moving window located at each possible position of I. We measured the similarity between the image and the template using the Normalized Cross Correlation (NCC). NCC is invariant to linear changes in brightness and/or contrast. NCC between template T and image I at pixel (x, y) is calculated as: This basic approach is not well suited for detecting junctions and terminals in the magnetic labyrinthine structures because a single template is not capable of modeling: All possible rotations;


All deformed shapes of defects. All possible rotations; All possible rotations; All deformed shapes of defects. All deformed shapes of defects. To solve problem (1), we employ a rotation-invariant template matching based on exhaustive evaluation of rotated templates. There are some alternative rotation-invariant techniques based on circular and radial projections [32], [34], and on Fourier coefficients of circular and radial projections [35]. These techniques can reduce computational requirements, but their implementations are complex and require parameter tuning. Furthermore, our application does not require exceptional computational performance, as the processing is offline. Thus, our technique uses the standard OpenCV[1] template matching implementation, which is highly optimized using FFT and special processor instructions. D. TEMPLATES AND MASKS USED IN THE EXPERIMENT D. TEMPLATES AND MASKS USED IN THE EXPERIMENT D. TEMPLATES AND MASKS USED IN THE EXPERIMENT We manually designed templates and masks, and empirically tuned them to capture various junction and terminal shapes. They have 21×21 resolutions and are generated at runtime. Templates represent magnetic strips as black lines radiating from their center drawn on a white background. Meanwhile, masks are created from black backgrounds with white areas defining relevant coordinates used in template matching. Their main purpose is to obscure the space between strips and background, addressing variations in widths and curvatures, and to limit the background regions to reduce interference from neighboring strips. Table 1 exemplifies the templates and masks used to detect junctions and terminals. Template matching is applied separately for each (template, mask) pair. Several template matchings are computed in parallel using the OpenMP library[2]. This process takes about 80 seconds to process an image on an i7-9750H processor. To obtain the final correlation map corr, we calculate the maximum value among all n = 3 × 439 + 5 × 120 = 1917 NCC maps for each position (x, y), that is: Pixels where corr(x, y) exceeds a predefined threshold t are considered potential detections. However, a single junction/terminal may encompass multiple neighboring points with correlation values greater than the threshold t. Therefore, it is necessary to perform some form of non-maximum suppression to eliminate duplicate detections and select only the true center of junction/terminal. Kim et al. [18] present a solution to this problem: Whenever two potential detection points p1 and p2 are separated by a distance smaller than a threshold c, the point with the lowest correlation value is discarded. In this work, we use a slightly different approach, but with the same practical result: Whenever the algorithm finds a potential detection, it executes a breadth-first search algorithm. This algorithm recursively searches adjacent pixels (x, y) where the correlation exceeds 80% of the threshold (that is, corr(x, y) > 0.8t) and saves the pixel with the highest correlation. Subsequently, the searched area has its correlation value set to zero to avoid re-detection. Fig. 3d highlights the searched area in red. The pixel with the highest correlation is chosen as the center of the junction/termination. This process performs detection in a single pass. E. DATASET ANNOTATION E. DATASET ANNOTATION E. DATASET ANNOTATION To classify potential detections into true or false using a CNN classifier, we must first create annotated training images. TM-CNN makes it easier to create training examples, as it allows one to annotate the examples semi-automatically. This process is divided in two phases. 1) Template matching-assisted annotation 1) Template matching-assisted annotation In this phase, only a small set of images are annotated. Initially, we apply template matching followed by nonmaximum suppression to identify the centers of possible detections, together with their probable labels (junction or terminal). Without this help from template matching, we would have to manually and precisely locate the centers of thousands of defects. Subsequently, a human reviewer makes corrections to ensure that the labels given by the template matching are correct. The reviewer may change the labels to junction, terminal or false detection. After all positive detections are annotated along with a small set of false detections, a larger set of false detections is created by lowering the template matching threshold and sampling new false detections. These images, now with positive and negative annotations, are used to train a preliminary version of the CNN classifier. 2) Deep learning-assisted annotation 2) Deep learning-assisted annotation Due to the small number of images in the initial training set, the preliminary CNN classifier cannot accurately classify all magnetic stripe defects. Nonetheless, this preliminary model is integrated into the annotation procedure to alleviate the required workload. In the second phase, we continue using template matching to generate the initial set of detections. However, the preliminary CNN classifier is employed to identify most of the template matching errors, thus speeding up the annotation process. As new images are annotated, more accurate models are trained to further simplify the annotation workload. The final training set consists of 17 images derived from a single annealing protocol, selected to cover varied experimental configurations of ascending and descending magnetic fields at different magnitudes. Out of these, 16 were selected from the quenched (unordered) state, as they represent a more diverse set of shapes and represent a greater challenge for classification. The training set encompasses a total of 33,772 detections, which includes 12,144 junctions, 12,777 terminals, and 8,851 false detections. F. CANDIDATE FILTERING BY CNN F. CANDIDATE FILTERING BY CNN F. CANDIDATE FILTERING BY CNN Our algorithm extracts small 50×50 patches centered around each detection point and a CNN classifies them into three classes: junction, terminal or false positive. The size of patches for CNN classification is larger than the size of template matching models (21×21), allowing CNN to use more contextual information than template matching. We use a simple CNN model to classify the small patches (Fig. 4). It has four convolutional layers with 32, 64, 128 and 256 filters, all using 3×3 kernels. The first three convolutional layers are followed by max pooling layers to downsample the feature maps and global max pooling is applied after the last convolutional layer. This is followed by dropout and two fully connected layers: the first with 128 nodes and the second with three output nodes. The ReLU activation function is used across the model, except at the output layer where the softmax is used. In total, this network has only 422,608 parameters. For comparison, VGG-16 and ResNet-50 (common backbones for detection) have 138 million and 23 million parameters, respectively. Thanks to its simplicity, our model is fast and can make predictions even without GPUs and takes around 30 seconds for filtering each image using an i7-9750H processor with 16 GB of RAM. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv [1] Open Source Computer Vision Library, https://opencv.org [2] Open Multi-Processing, https://www.openmp.org/ [3] https://www.tensorflow.org/

Why TM-CNN Is a Material Scientist's New Best Friend

Scientists Fuse Human Brain Cells With Chips: Is This the Birth of Living AI?

What is custom transformation logic? Index-as-a-service (or RAG-as-service), tends to package a predesigned service and expose two endpoints to users - one to configure the source, and an API to read from the index. Many predefined pipelines for unstructured documents do this. The requirements are fairly simple: parse PDFs, perform some chunking and embedding, and dump into vector stores. This works well if your requirements are simple and primarily focused on document parsing. We've talked to many developers across various verticals that require data indexing, and being able to customize logic is essential for high-quality data retrieval. For example: Basic choices for pipeline components

which parser for different files?
how to chunk files (documents with structure normally have different optimal chunking strategies)?
which embedding model? which vector database?



What should the pipeline do?

Is it a simple text embedding"?
Is it building a knowledge graph?
Should it perform simple summarization for each source for retrieval without chunking?



What additional work is needed to improve pipeline quality?

Do we need deduplication?
Do we need to look up different sources to enrich our data?
Do we need to reconcile and align multiple documents? Basic choices for pipeline components

which parser for different files?
how to chunk files (documents with structure normally have different optimal chunking strategies)?
which embedding model? which vector database? Basic choices for pipeline components which parser for different files?
how to chunk files (documents with structure normally have different optimal chunking strategies)?
which embedding model? which vector database? which parser for different files? how to chunk files (documents with structure normally have different optimal chunking strategies)? which embedding model? which vector database? What should the pipeline do?

Is it a simple text embedding"?
Is it building a knowledge graph?
Should it perform simple summarization for each source for retrieval without chunking? What should the pipeline do? Is it a simple text embedding"?
Is it building a knowledge graph?
Should it perform simple summarization for each source for retrieval without chunking? Is it a simple text embedding"? Is it building a knowledge graph? Should it perform simple summarization for each source for retrieval without chunking? What additional work is needed to improve pipeline quality?

Do we need deduplication?
Do we need to look up different sources to enrich our data?
Do we need to reconcile and align multiple documents? What additional work is needed to improve pipeline quality? Do we need deduplication?
Do we need to look up different sources to enrich our data?
Do we need to reconcile and align multiple documents? Do we need deduplication? Do we need to look up different sources to enrich our data? Do we need to reconcile and align multiple documents? Here we'll walk through some examples of the topology of index pipelines, and we can explore more in the future! Basic embedding​ ​ In this example, we do the following: In this example, we do the following: Read from sources, for example, a list of PDFs
For each source file, parse it to markdown with a PDF parser. There are lots of choices out there: Llama Parse, Unstructured IO, Gemini, DeepSeek, etc.
Chunk all the markdown files. This is a way to break text into smaller chunks, or units, to help organize and process information. There are many options here: flat chunks, hierarchical chunks, secondary chunks, and many publications in this area. There are also special chunking strategies for different verticals - for code, tools like Tree-sitter can help parse and chunk based on syntax. Normally the best choice is tied to your document structure and requirements.
Perform embedding for each chunk. There are lots of great choices: Voyage, models like OpenAI etc.
Collect the embeddings in vector stores. The embedding is normally attached with metadata, for example, which file this embedding belongs to, etc. There are many great choices for vector stores: Chromadb, Milvus, Pinecone and many databases now support vector indexing, for example PostgreSQL (pgvector) and MongoDB. Read from sources, for example, a list of PDFs For each source file, parse it to markdown with a PDF parser. There are lots of choices out there: Llama Parse, Unstructured IO, Gemini, DeepSeek, etc. Llama Parse Unstructured IO Gemini DeepSeek Chunk all the markdown files. This is a way to break text into smaller chunks, or units, to help organize and process information. There are many options here: flat chunks, hierarchical chunks, secondary chunks, and many publications in this area. There are also special chunking strategies for different verticals - for code, tools like Tree-sitter can help parse and chunk based on syntax. Normally the best choice is tied to your document structure and requirements. Tree-sitter Perform embedding for each chunk. There are lots of great choices: Voyage, models like OpenAI etc. Voyage OpenAI Collect the embeddings in vector stores. The embedding is normally attached with metadata, for example, which file this embedding belongs to, etc. There are many great choices for vector stores: Chromadb, Milvus, Pinecone and many databases now support vector indexing, for example PostgreSQL (pgvector) and MongoDB. Chromadb Milvus Pinecone A combination of TF-IDF and vector search​ ​ Anthropic has published a great article about Contextual Retrieval that suggests the combination of vector-based search and TF-IDF. Contextual Retrieval The way to think about a data flow for the pipeline is: In addition to prepare the vector embedding as basic embedding example above, after the source data parsing, we can do the following: For each document, extract keywords from it with its frequency.
Across all documents, group by keywords and sum up their frequencies, putting into an internal storage.
For each document each keyword, calculate the TF-IDF score using two inputs: frequency in the current document, and the sum of frequency across all documents. Store keyword and along with TF-IDF score keyword (if above a certain threshold) to a keyword index. For each document, extract keywords from it with its frequency. Across all documents, group by keywords and sum up their frequencies, putting into an internal storage. For each document each keyword, calculate the TF-IDF score using two inputs: frequency in the current document, and the sum of frequency across all documents. Store keyword and along with TF-IDF score keyword (if above a certain threshold) to a keyword index. On query time, we can query both the vector index and the keyword index, and combine results from both. Simple Data Lookup/Enrichment example​ ​ Sometimes, you want to enrich your data with metadata looked up from other sources. For example, if we want to create index on diagnostic reports, which uses ICD-10 (International Classification of Diseases Version 10) codes to describe diseases, we can have a pipeline like this: ICD-10 In this example, we do the following: On the first path, build a ICD-10 dictionary by

For each ICD-10 description document, convert it to markdown with a PDF parser.
Split into items.
For each item, extract the ICD-10 code and description, and collect into a storage.



On the second path, for each report

Parse it to markdown with a PDF parser.
Split into items.
For each item, lookup the ICD-10 dictionary prepared above and enrich the item with the descritions for ICD-10 codes. On the first path, build a ICD-10 dictionary by

For each ICD-10 description document, convert it to markdown with a PDF parser.
Split into items.
For each item, extract the ICD-10 code and description, and collect into a storage. On the first path, build a ICD-10 dictionary by For each ICD-10 description document, convert it to markdown with a PDF parser.
Split into items.
For each item, extract the ICD-10 code and description, and collect into a storage. For each ICD-10 description document, convert it to markdown with a PDF parser. Split into items. For each item, extract the ICD-10 code and description, and collect into a storage. On the second path, for each report

Parse it to markdown with a PDF parser.
Split into items.
For each item, lookup the ICD-10 dictionary prepared above and enrich the item with the descritions for ICD-10 codes. On the second path, for each report Parse it to markdown with a PDF parser.
Split into items.
For each item, lookup the ICD-10 dictionary prepared above and enrich the item with the descritions for ICD-10 codes. Parse it to markdown with a PDF parser. Split into items. For each item, lookup the ICD-10 dictionary prepared above and enrich the item with the descritions for ICD-10 codes. Now we have a vector index, built based on diagnostic reports enriched with ICD-10 descriptions. CocoIndex is an open-source engine that supports both custom transformation logic and incremental updates specialized for data indexing.  It supports various builtins to simplify your needs to customize data pipeline for AI.   ❤️ CocoIndex is Open Sourced under Apache License 2.0. If you like our work, please support us with a Github Star ⭐ at https://github.com/cocoindex-io/cocoindex. CocoIndex is an open-source engine that supports both custom transformation logic and incremental updates specialized for data indexing.  It supports various builtins to simplify your needs to customize data pipeline for AI.   ❤️ CocoIndex is Open Sourced under Apache License 2.0. If you like our work, please support us with a Github Star ⭐ at https://github.com/cocoindex-io/cocoindex. CocoIndex Apache License 2.0 https://github.com/cocoindex-io/cocoindex

Stories, News & Mentions about AIREDEFINED

Airedefined’s Stories on HackerNoon

Airedefined's News & Mentions