Tech News

Meta Learning, an original concept of cognitive psychology, is now applied to machine learning techniques. If we go by the social psychology definition, meta learning is the state of being aware of and taking control of one's own learning. Similar concepts, when applied to the machine learning theory states that a meta learning algorithm uses prior experience to change certain aspects of an algorithm, such that the modified algorithm is better than the original algorithm. To explain in simple terms, meta-learning is how the algorithm learns how to learn. Meta Learning: Making a versatile AI agent Current AI Systems excel at mastering a single skill, playing Go, holding human-like conversations, predicting a disaster, etc. However, now that AI and machine learning is possibly being integrated in everyday tasks, we need a single AI system to solve a variety of problems. Currently, a Go Player, will not be able to navigate the roads or find new places. Or an AI navigation controller won’t be able to hold a perfect human-like conversation. What machine learning algorithms need to do is develop versatility – the capability of doing many different things. Versatility is achieved by intelligent amalgamation of Meta Learning along with related techniques such as reinforcement learning (finding suitable actions to maximize a reward), transfer learning (re-purposing a trained model for a specific task on a second related task), and active learning (learning algorithm chooses the data it wants to learn from). Such different learning techniques provides an AI agent with the brains to do multiple tasks without the need to learn every new task from scratch. Thereby making it capable of adapting intelligently to a wide variety of new, unseen situations. Apart from creating versatile agents, recent researches also focus on using meta learning for hyperparameter and neural network optimization, fast reinforcement learning, finding good network architectures and for specific cases such as few-shot image recognition. Using Meta Learning, AI agents learn how to learn new tasks by reusing prior experience, rather than examining each new task in isolation. Various approaches to Meta Learning algorithms A wide variety of approaches come under the umbrella of Meta-Learning. Let's have a quick glance at these algorithms and techniques: Algorithm Learning (selection) Algorithm selection or learning, selects learning algorithms on the basis of characteristics of the instance. For example, you have a set of ML algos (Random Forest, SVM, DNN), data sets as the instances and the error rate as the cost metric. Now, the goal of Algorithm Selection is to predict which machine learning algorithm will have a small error on each data set. Hyper-parameter Optimization Many machine learning algorithms have numerous hyper-parameters that can be optimized. The choice of selecting these hyper-parameters for learning algorithms determines how well the algorithm learns.  A recent paper, "Evolving Deep Neural Networks", provides a meta learning algorithm for optimizing deep learning architectures through evolution. Ensemble Methods Ensemble methods usually combine several models or approaches to achieve better predictive performance. There are 3 basic types – Bagging, Boosting, and Stacked Generalization. In Bagging, each model runs independently and then aggregates the outputs at the end without preference to any model. Boosting refers to a group of algorithms that utilize weighted averages to make weak learners into stronger learners. Boosting is all about “teamwork”. Stacked generalization, has a layered architecture. Each set of base-classifiers is trained on a dataset. Successive layers receive as input the predictions of the immediately preceding layer and the output is passed on to the next layer. A single classifier at the topmost level produces the final prediction. Dynamic bias selection In Dynamic Bias selection, we adjust the bias of the learning algorithm dynamically to suit the new problem instance. The performance of a base learner can trigger the need to explore additional hypothesis spaces, normally through small variations of the current hypothesis space. The bias selection can either be a form of data variation or a time-dependent feature. Inductive Transfer Inductive transfer describes learning using previous knowledge from related tasks. This is done by transferring meta-knowledge across domains or tasks; a process known as inductive transfer. The goal here is to incorporate the meta-knowledge into the new learning task rather than matching meta-features with a meta-knowledge base. Adding Enhancements to Meta Learning algorithms Supervised meta-learning:  When the meta-learner is trained with supervised learning. In supervised learning we have both input and output variables and the algorithm learns the mapping function from the input to the output. RL meta-learning: This algorithm talks about using standard deep RL techniques to train a recurrent neural network in such a way that the recurrent network can then implement its own Reinforcement learning procedure. Model-agnostic meta-learning: MAML trains over a wide range of tasks, for a representation that can be quickly adapted to a new task, via a few gradient steps. The meta-learner seeks an initialization that is not only useful for adapting to various problems, but also can be adapted quickly. The ultimate goal of any meta learning algorithm and its variations is to be fully self-referential. This means it can automatically inspect and improve every part of its own code. A regenerative meta learning algorithm, on the lines of how a lizard regenerates its limbs, would not only blur the distinction between the different variations as described above but will also lead to better future performance and versatility of machine learning algorithms.

In October last year, it was reported that LLVM (Low-Level Virtual Machine) is moving from Apache Subversion (SVN) to GitHub. Now the migration is complete and LLVM is available on GitHub. This transition was long under discussion and is now officially complete. LLVM is a toolkit for creating compilers, optimizers, and runtime environments. This migration comes in place as continuous integration is sometimes broken in LLVM because the SVN server was down. They migrated to GitHub for services lacking in SVN such as better 24/7 stability, disk space, code browsing, forking etc. GitHub is also used by most of the LLVM community. There already were unofficial mirrors on GitHub before this official migration. Last week, James Y Knight from the LLVM team wrote to a mailing list: “The new official monorepo is published to LLVM's GitHub organization, at: https://github.com/llvm/llvm-project. At this point, the repository should be considered stable -- there won't be any more rewrites which invalidate commit hashes (barring some _REALLY_ good reason...)” Along with LLVM, this monorepo also hosts Clang, LLD, Compiler-RT, and other LLVM sub-projects. Commits are being made to the LLVM GitHub repository even at the time of writing and the repo currently has about 200 stars. Updated workflow documents and instructions on migrating user work that is in-progress are being drafted and will be available soon. This move was initiated after positive responses from LLVM community members to migrate to GitHub. If you want to be up to date with more details, you can follow the LLVM mailing list. LLVM will be relicensing under Apache 2.0 start of next year A libre GPU effort based on RISC-V, Rust, LLVM and Vulkan by the developer of an earth-friendly computer LLVM 7.0.0 released with improved optimization and new tools for monitoring

Yesterday, ZDNet reported that the Python security team removed two fake Python libraries from PyPI (Python Package Index). These libraries were caught stealing SSH and GPG keys from the Python projects. As per ZDNet, the two malicious clones were discovered by a German software developer Lukas Martini on 1st Dec. Both libraries were removed on the same day after Martini notified the developers and the PyPI security team. The two libraries were created by the same developer and mimicked as other more popular libraries -- using a technique called typosquatting, to register similar-looking names. The first is "python3-dateutil," which imitated the popular "dateutil" library. The second is "jeIlyfish" (here the first L is an I), which mimicked the "jellyfish" library. One of them was uploaded on Pypi two days before while the other one was live for more than a year. Purpose of stealing SSH and GPG keys According to Martini, the malicious code was present only in the jeIlyfish library. The python3-dateutil package didn't contain malicious code of its own, but it did import the jeIlyfish library, meaning it was malicious by association. The malicious code read a list of hashes stored in a GitLab repository. The nature and purpose of these hashes is unknown, as neither Martini or the PyPI team detailed the behavior of stealing the keys before the library was removed. ZDNet spoke to Paul Ganssle from the dateutil dev team, "The code directly in the `jeIlyfish` library downloads a file called 'hashsum' that looks like nonsense from a gitlab repo, then decodes that into a Python file and executes it," Ganssle states. "It looks like [this file] tries to exfiltrate SSH and GPG keys from a user's computer and sends them to this IP address: http://68.183.212.246:32258. It also lists a bunch of directories, home directory, PyCharm Projects directory," Ganssle added. "If I had to guess what the purpose of that is, I would say it's to figure out what projects the credentials work for so that the attacker can compromise that person's projects." Python developers advised to review projects Excluding the malicious code, both typosquatted packages were identical copies of the original libraries, meaning they would have worked as the originals. Developers who didn't pay attention to the libraries they downloaded or imported into their projects are advised to check if they've used the correct package names and did not accidentally use the typosquatted versions. If they accidentally used any of the two, developers must change all SSH and GPG keys which they've used over the past year. This is the third time the PyPI team intervenes to remove typo-squatted malicious Python libraries from the official repository. Similar incidents took place recently in July 2019 and another in October 2018 and September 2017. On this news, developers on Hacker News discuss about this as an OS issue. One of the user comments, “I don't know what the solution is but it feels like this is a much bigger issue and we need some rethinking of how OSes work by default. Apple has taken some steps it seems the last 2 MacOS updates where they block access to certain folders for lots of executables until the user specifically gives that permission. Unfortunately for things like python the permission is granted to the Terminal app so once given, all programs running under the terminal inherit the permissions. Microsoft has started adding short life VMs. No idea if that's good. Both MS and Apple offer their App stores with more locked down experiences though I'm sad they conflate app security and app markets. Basically anytime I run any software, everytime I run "make" or "npm install" or "pip install" or download a game on Steam etc I'm having to trust 1000s of strangers they aren't downloading my keys, my photos, my docs, etc...I think you should be in control of your machine but IMO it's time to default to locked down instead of defaulting to open.” Introducing Spleeter, a Tensorflow based python library that extracts voice and sound from any music track SatPy 0.10.0, python library for manipulating meteorological remote sensing data, released Meet Pypeline, a simple python library for building concurrent data pipelines

Last week, the team behind Inkscape project released the first beta version of the upcoming and much-awaited Inkscape 1.0. The team writes on the announcement page that, “after releasing two less visible alpha versions this year, in mid-January and mid-June (and one short-lived beta version), Inkscape is now ready for extensive testing and subsequent bug-fixing.” Most notable changes in Inkscape 1.0 New theme selection: In 'Edit > Preferences > User Interface > Theme', users can set a custom GTK3 theme for Inkscape. If the theme comes with a dark variant, activating the 'Use dark theme' checkbox will result in the dark variant being used. Then the new theme will be applied immediately. Origin in top left corner: Another significant change integrated was to set the origin of the document to the top left corner of the page. It coordinates that a user can see in the interface match the ones that are saved in the SVG data, and makes working in Inkscape more comfortable for people who are used to this standard behavior. Canvas rotation and mirroring: With Ctrl+Shift+Scroll wheel the drawing area can be rotated and viewed from different angles. The canvas can be flipped, to ensure that the drawing does not lean to one side, and looks good either way. Canvas alignment: When the option "Enable on-canvas alignment" is active in the "Align and Distribute" dialog, a new set of handles will appear on canvas. These handles can be used to align the selected objects relative to the area of the current selection. HiDPI screen: Inkscape now supports HiDPI screens. Controlling PowerStroke: The width of PowerStroke is controlled with pressure sensitive graphics tablet Fillet/chamfer LPE and (non-destructive) Boolean Operation LPE: This new LPE adds fillet and chamfer to paths. The Boolean Operations LPE finally makes non-destructive boolean operations available in Inkscape. New PNG export options: The export dialog has received several new options available when you expand the 'Advanced' section. Centerline tracing: A new, unified dialog for vectorizing raster graphics is now available from Path > Trace Bitmap. New Live Path Effect selection dialog: Live Path Effects received a major overhaul, with lots of improvements and new features. Faster Path operations and deselection of large number of paths Variable fonts support: If Inkscape is compiled with a Pango library version 1.41.1, then it will come with support for variable fonts. Complete extensions overhaul: Extensions can now have clickable links, images, a better layout with separators and indentation, multiline text fields, file chooser fields and more. Command line syntax changes: The Inkscape command line is now more powerful and flexible for the user and easier to enhance for the developer. Native support for macOS with a signed and notarized .dmg file: Inkscape is now a first-rate native macOS application, and no longer requires XQuartz to operate. Other important changes for users Custom Icon Sets Icon sets no longer consist of a single file containing all icons. Instead each icon is allocated it's own file. The directory structure must follow the standard structure for Gnome icons. As a side effect of a bug fix to the icon preview dialog, custom UI icon SVG files need to be updated to have their background color alpha channel set to 0 so that they display correctly. Third-party extensions Third-party extensions need to be updated to work with this version of Inkscape. Import/Export via UniConvertor dropped Extensions that previously used the UniConvertor library for saving/opening various file formats have been removed: Import formats that have been removed: Adobe Illustrator 8.0 and below (UC) (*.ai) Corel DRAW Compressed Exchange files (UC) (*.ccx) Corel DRAW 7-X4 files (UC) (*.cdr) [cdr imports, but this specific version?] Corel DRAW 7-13 template files (UC) (*.cdt) Computer Graphics Metafile files (UC) (*.cgm) Corel DRAW Presentation Exchange files (UC) (*.cmx) HP Graphics Language Plot file [AutoCAD] (UC) (*.plt) sK1 vector graphics files (UC) (*.sk1) Export formats that have been removed: HP Graphics Language Plot file [AutoCAD] (UC) (*.plt) sK1 vector graphics files (UC) (*.sk1) Inline LaTeX formula conversion dropped The EQTeXSVG extension that could be used to convert an inline LaTeX equation into SVG paths using Python was dropped, due to its external dependencies. The team has asked to test Inkscape 1.0 beta version and report the findings on Inkscape report page. To know more about this news, check out official Inkscape announcement page. Other interesting news in web development this week! Mozilla announces final four candidates that will replace its IRC network Announcing Feathers 4, a framework for real-time apps and REST APIs with JavaScript or TypeScript Google announces two new attribute links, Sponsored and UGC and updates “nofollow”

Microsoft is ever striving to make its products run better. They can add yet another accomplishment to their list as Microsoft’s cloud service search engine, Bing is now running fully on .NET Core 2.1, as announced by the .NET engineering team in their blog yesterday. .NET Core is the slimmed down and cross-platform version of Microsoft’s .NET managed common language runtime. Since Bing runs on thousands of servers spanning many data centers across the globe, .NET Core will serve as the perfect platform for it to function on. Why did Bing migrate to .NET Core 2.1? Bing has always run on the .NET Framework, but has been able to move to .NET Core 2.1 after some recent API additions. Let’s take a look at the main reasons for Bing.com’s migration to .NET Core. 1. Performance i.e. serving latency .NET Core 2.1 has led to an improvement in performance in virtually all areas of the runtime and libraries. The internal server latency over the last few months shows a striking 34% improvement. Check out the graph for a clear picture!     Souce: blog.msdn.microsoft.com The following changes in .NET Core 2.1 are the reasons why the workload and performance has greatly improved- #1 Vectorization of string.Equals & string.IndexOf/LastIndexOf HTML rendering and manipulation are string-heavy workloads. Vectorization of String comparisons and indexing operations (major components of string slicing) is the biggest contributor to the performance improvement. You can find more information on this on the github page for  Vectorization of string.Equals and string.IndexOf/LastIndexOf #2 Devirtualization Support for EqualityComparer<T>.Default One of .NET core’s major software components is a heavy user of Dictionary<int/long, V>, which indirectly benefits from the intrinsic recognition work that was done in the JIT to make Dictionary<K, V> amenable to that optimization.  Head over to the github page for more clarity on why this feature empowers .NET Core 2.1 #3 Software Write Watch for Concurrent GC This led to a reduction in CPU usage. The implementation relies on a JIT Write Barrier, which instinctively increases the cost of a reference store, but that cost is amortized and not noticed in the workload. #4 Methods with calli are now inline-able ldftn + calli  are used in lieu of delegates (which incur an object allocation) in performance-critical pieces of code where there is a need to call a managed method indirectly. This change allowed method bodies with a calli instruction to be eligible for inlining. The github page provides more insight on this subject. #5 Improve performance of string.IndexOfAny for 2 & 3 char searches A common operation in a front-end stack is search for ‘:’, ‘/’, ‘/’ in a string to delimit portions of a URL. Check out this special-casing improvement that was beneficial throughout the codebase on the github page. 2. Runtime Agility The ability to have an xcopy version of the runtime inside their application denotes that they can adopt newer versions of the runtime at a much faster pace. The Continuous integration (CI) pipeline is run with .NET Core’s daily CI and it builds testing functionality and performance all the way through the release. 3. ReadyToRun Images Managed applications usually can have poor startup performance as methods first have to be JIT compiled to machine code. .NET Framework has a precompilation technology, NGEN. On .NET Core, the crossgen tool allows the code to be precompiled as a pre-deployment step, such as in the build lab, and the images deployed to production are Ready To Run! This feature was not supported on the previous  .NET implementation. The .NET Core team is striving to provide Bing.com users fast results. The latest software and technologies used by their developers will ensure that .NET Core will not fail Bing.com! Read the detailed overview of the article on Microsoft's blog. Say hello to FASTER: a new key-value store for large state management by Microsoft Microsoft Azure’s new governance DApp: An enterprise blockchain without mining .NET Core completes move to the new compiler – RyuJIT

Yesterday, Microsoft announced the launch of the first beta builds of its Chromium-based Edge browser. Developers using the supported versions of Windows and macOS can now test it and share their feedback with the Microsoft Edge team. https://twitter.com/MicrosoftEdge/status/1163874455690645504 The preview builds are made available to developers and early adopters through three channels on the Microsoft Edge Insider site: Beta, Canary, and Developer. Earlier this year, Microsoft opened the Canary and Developer channel. Canary builds receive updates every night, while Developer builds are updated weekly. Beta is the third and final preview channel that is the most stable one among the three. It receives updates every six weeks, along with periodic minor updates for bug fixes and security. Source: Microsoft Edge Insider What’s new in Microsoft Edge Beta Microsoft Edge Beta comes with several options using which you can personalize your browsing experience. It supports the dark theme and offers 14 different languages to choose from. If you are not a fan of the default new tab page, you can customize it with its tab page customizations. There are currently three preset styles that you can switch between:  Focused, Inspirational and Informational. You can further customize and personalize Microsoft Edge through the different addons available on Microsoft Edge Insider Addons store or other Chromium-based web stores. Talking about the user privacy bit, this release brings support for tracking prevention. Enabling this feature will protect you from being tracked by websites that you don’t visit. You can choose from three levels of privacy, which are Basic, Balanced and Strict. Microsoft Edge Beta also comes with some of the commercial features that were announced at Build this year. Microsoft Search is now integrated into Bing that lets you search for OneDrive files directly from Bing search. There is support for Internet Explorer mode that brings Internet Explorer 11 compatibility directly into Microsoft Edge. It also supports Windows Defender Application Guard for isolating enterprise-defined untrusted sites. Along with this release, Microsoft also launched the Microsoft Edge Insider Bounty Program.  Under this program researchers who find any high-impact vulnerabilities in Dev and Beta channels will receive rewards up to US$30,000. Read Microsoft’s official announcement to know more in detail. Microsoft officially releases Microsoft Edge canary builds for macOS users Microsoft Edge mobile browser now shows warnings against fake news using NewsGuard Microsoft Edge Beta available on iOS with a breaking news alert, developer options and more

The highly anticipated Unreal Engine 4.19 is now generally available. This release hosts a new Live Link plugin, improvements to Sequencer, new Dynamic Resolution feature, and multiple workflow and usability improvements. In addition to all of these major updates, this release also features a massive 128 improvements based on queries submitted by the Unreal Engine developers community on GitHub. Unreal Engine 4.19 allows game developers to know exactly what their finished game will look like at every step of the development process. This update comes with three major goals: Let developers step inside the creative process. Build gaming worlds that run faster than ever before. Give developers full control. Here's a list of all the major features and what they bring to the game development process: New Unreal Engine 4.19 features Live Link Plugin Improvements The Maya Live Link Plugin is now available and can be used to establish a connection between Maya and UE4 to preview changes in real-time. Virtual Subjects are added to the Live Link. It can also be used with Motion Controllers. Live Link Sources can now define their own custom settings. Virtual Initialization function and Update DeltaTime parameter are also added to Live Link Retargeter API. Source: Unreal Engine blog Unified Unreal AR framework The Unreal Augmented Reality Framework provides a unified framework for building Augmented Reality (AR) apps for both Apple and Google handheld platforms using a single code path. Features include functions supporting Alignment, Light Estimation, Pinning, Session State, Trace Results, and Tracking. Temporal upsampling The new upscaling method, Temporal Upsample performs two separate screen percentages used for upscaling: Primary screen percentage that by default will use the spatial upscale pass as before; Secondary screen percentage that is a static, spatial only upscale at the very end of post-processing, before the UI draws. Dynamic resolution Dynamic Resolution adjusts the resolution to achieve the desired frame rate, for games on PlayStation 4 and Xbox One. It uses a heuristic to set the primary screen percentage based on the previous frames GPU workload. Source: Unreal Engine blog Physical light units All light units are now defined using physically based units. The new light unit property can be edited per light, changing how the engine interprets the intensity property when doing lighting related computations. Source: Unreal Engine blog Landscape rendering optimization The Landscape level of detail (LOD) system now uses screen size to determine detail for a component, similar to how the Static Mesh LOD system works. Starting from this release, all existing UE4 content that supports SteamVR is now compatible with HTC's newly-announced Vive Pro. These are just a select few updates to the Unreal Engine. The full list of release notes is available on the Unreal Engine forums.

Three days ago, the team behind Python announced the release of Python 3.9.0a1, which is the first out of the six planned alpha releases of Python 3.9. The final stable version of Python 3.9 is slated to release in May 2020. An alpha release indicates that developers can start testing the new features and check for bug fixes but are not recommended to use it in production. Last month, the previous stable version, Python 3.8 was released with features like walrus operator, positional-only parameters support for Vectorcall. Read More: Core Python team confirms sunsetting Python 2 on January 1, 2020 Let’s look at some of the raw features that you can be expected in the upcoming Python 3.9 version. Some improvements introduced in Python 3.9.0a1 Language Changes The __import__() function which is invoked by the import statement will now raise ImportError instead of ValueError. In the previous versions, the latter used to occur when a relative import went past its top-level package. Starting from Python 3.9.0a1, the absolute path of the script filename will be specified on the command line: the __file__ attribute of the __main__ module. The sys.argv[0] and sys.path[0] will become an absolute path rather than a relative path. Also, the traceback will now display the absolute path for __main__ module frames in this case. The encoding and errors arguments in the debug build and development mode will now be checked in the string encoding and decoding operations. Improved Modules ast: It is added in the indent option to dump() and produces a multi-line indented output. asyncio: It can now use coroutine which is a generalized form of subroutines. Subroutines enter and exit at only two different points, while coroutines can be entered, exited, and resumed at many points. Moreover, asyncio.run() is updated to use the new coroutine. New functions like curses.get_escdelay(), curses.set_escdelay(), curses.get_tabsize(), and curses.set_tabsize() and constants F_OFD_GETLK, F_OFD_SETLK and F_OFD_SETLKW is included in Python 3.9.0a1. Few Python users have already started testing the Python 3.9.0a1 release. https://twitter.com/codewithanthony/status/1197559895744110592 The next alpha release for Python 3.9 is scheduled for 16th December 2019. To know more about Python 3.9.0a1, check out the official documentation. Introducing Spleeter, a Tensorflow based python library that extracts voice and sound from any music track Severity issues raised for Python 2 Debian packages for not supporting Python 3 Introducing OpenDrop, an open-source implementation of Apple AirDrop written in Python Poetry, a Python dependency management and packaging tool, releases v1 beta 1 with URL dependency PyPy will continue to support Python 2.7, even as major Python projects migrate to Python 3

Qml.Net is a C# library for cross-platform GUI development with native dependency. It exposes the required object types to host a QML engine. In Qml.NET, Qml and JavaScript together form the UI layer. It can be thought of as the view in MVC. Qml.Net features The PInvoke code in this .NET library is hand-crafted by developer Paul Knopf to ensure appropriate memory management and pointer ownership semantics. He is pretty confident about the library and mentions in his blog “I’d bet you couldn’t generate a segfault, even if you wanted to.” In Qml.Net C# objects can be registered to be treated as QML components. You can then interoperate with them as you would with regular JavaScript objects. The registered C# objects serve as a portal through which the QML world can interact with your .NET objects. This has an added benefit of keeping your business/UI concerns separate cleanly. There will also be no chatty PInvoke calls for rendering. It is a great match. A pre-compiled portable installation of Qt and the native C wrapper is available for Windows, OSX, and Linux. Developers wouldn’t have to bother with C/C++. All you need to know is QML, C#, and JavaScript; QML if fairly simple. QML can’t really be classified as a language, in the semantic sense. More appropriately it can be considered as a combination of JSON and JavaScript. Qml.Net support and working Qml.Net will work with any .NET language including popular C# and functional languages like F#. Your libraries will reference the pure .NET NuGet package, Qml.Net. The host process (Program.Main) references the native NuGet packages. This is dependent on the OS you are on: Qml.Net.WindowsBinaries Qml.Net.OSXBinaries Qml.Net.LinuxBinaries Paul currently only tests his own models that are C# objects registered with the QML engine. They are specific to each control/page. Since Microsoft's announcement of .NET Core, there hasn’t been any clear idea on cross-platform GUI development. Although Microsoft plans to support WPF in .NET Core 3.0, it will be limited to Windows machines. With community involvement and support, Qml.net can be a potential game changer. You can head to the GitHub repository and also view some hosted examples to get a better idea. Read next Exciting New Features in C# 8.0 .NET Core completes move to the new compiler – RyuJIT Microsoft Azure's new governance DApp: An enterprise blockchain without mining

Researchers from Facebook and Carnegie Mellon University have developed an AI bot that has defeated human professionals in six-player no-limit Texas Hold’em poker.   Pluribus defeated pro players in both “five AIs + one human player” format and a “one AI + five human players” format. Pluribus was tested in 10,000 games against five human players, as well as in 10,000 rounds where five copies of the AI  played against one professional. This is the first time an AI bot has beaten top human players in a complex game with more than two players or two teams. Pluribus was developed by Noam Brown of Facebook AI Research and Tuomas Sandholm of Carnegie Mellon University. Pluribus builds on Libratus, their previous poker-playing AI which defeated professionals at Heads-Up Texas Hold ’Em, a two-player game in 2017. Mastering 6-player Poker for AI bots is difficult considering the number of possible actions. First, obviously since this involves six players, the games have a lot more variables and the bot can’t figure out a perfect strategy for each game - as it would do for a two player game. Second, Poker involves hidden information, in which a player only has access to the cards that they see. AI has to take into account how it would act with different cards so it isn’t obvious when it has a good hand. Brown wrote on a Hacker News thread, “So much of early AI research was focused on beating humans at chess and later Go. But those techniques don't directly carry over to an imperfect-information game like poker. The challenge of hidden information was kind of neglected by the AI community. This line of research really has its origins in the game theory community actually (which is why the notation is completely different from reinforcement learning). Fortunately, these techniques now work really really well for poker.” What went behind Pluribus? Initially, Pluribus engages in self-play by playing against copies of itself, without any data from human or prior AI play used as input. The AI starts from scratch by playing randomly, and gradually improves as it determines which actions, and which probability distribution over those actions, lead to better outcomes against earlier versions of its strategy. Pluribus’s self-play produces a strategy for the entire game offline, called the blueprint strategy. This online search algorithm can efficiently evaluate its options by searching just a few moves ahead rather than only to the end of the game. Pluribus improves upon the blueprint strategy by searching for a better strategy in real time for the situations it finds itself in during the game. Real-time search The blueprint strategy in Pluribus was computed using a variant of counterfactual regret minimization (CFR). The researchers used Monte Carlo CFR (MCCFR) that samples actions in the game tree rather than traversing the entire game tree on each iteration. Pluribus only plays according to this blueprint strategy in the first betting round (of four), where the number of decision points is small enough that the blueprint strategy can afford to not use information abstraction and have a lot of actions in the action abstraction. After the first round, Pluribus instead conducts a real-time search to determine a better, finer-grained strategy for the current situation it is in. https://youtu.be/BDF528wSKl8 What is astonishing is that Pluribus uses very little processing power and memory, less than $150 worth of cloud computing resources. The researchers trained the blueprint strategy for Pluribus in eight days on a 64-core server and required less than 512 GB of RAM. No GPUs were used. Stassa Patsantzis, a Ph.D. research student appreciated Pluribus’s resource-friendly compute power. She commented on Hacker News, “That's the best part in all of this. I'm hoping that there is going to be more of this kind of result, signaling a shift away from Big Data and huge compute and towards well-designed and efficient algorithms.” She also said how this is significantly lesser than ML algorithms used at DeepMind and Open AI. “In fact, I kind of expect it. The harder it gets to do the kind of machine learning that only large groups like DeepMind and OpenAI can do, the more smaller teams will push the other way and find ways to keep making progress cheaply and efficiently”, she added. Real-life implications AI bots such as Pluribus give a better understanding of how to build general AI that can cope with multi-agent environments, both with other AI agents and with humans. A six-player AI bot has better implications in reality because two-player zero-sum interactions (in which one player wins and one player loses) are common in recreational games, but they are very rare in real life.  These AI bots can be used for handling harmful content, dealing with cybersecurity challenges, or managing an online auction or navigating traffic, all of which involve multiple actors and/or hidden information. Apart from fighting online harm, four-time World Poker Tour title holder Darren Elias helped test the program's skills, said, Pluribus could spell the end of high-stakes online poker. "I don't think many people will play online poker for a lot of money when they know that this type of software might be out there and people could use it to play against them for money." Poker sites are actively working to detect and root out possible bots. Brown, Pluribus' developer, on the other hand, is optimistic. He says it's exciting that a bot could teach humans new strategies and ultimately improve the game. "I think those strategies are going to start penetrating the poker community and really change the way professional poker is played," he said. For more information on Pluribus and it’s working, read Facebook’s blog. DeepMind’s Alphastar AI agent will soon anonymously play with European StarCraft II players Google DeepMind’s AI AlphaStar beats StarCraft II pros TLO and MaNa OpenAI Five bots destroyed human Dota 2 players this weekend

Yesterday, the Unity team announced the release of Unity 2019.2. In this release, they have added more than 170 new features and enhancements for artists, designers, and programmers. They have updated ProBuilder, Shader Graph, 2D Animation, Burst Compiler, UI Elements, and many more. Major highlights for Unity 2019.2 ProBuilder 4.0 ships as verified with 2019.2. It is a unique hybrid of 3D modeling and level design tools, optimized for building simple geometry but capable of detailed editing and UV unwrapping as needed. Polybrush is now available via Package Manager as a Preview package. This versatile tool lets you sculpt complex shapes from any 3D model, position detail meshes, paint in custom lighting or coloring, and blend textures across meshes directly in the Editor. DSPGraph is the new audio rendering/mixing system, built on top of Unity’s C# Job System. It’s now available as a Preview package. They have improved on UI Elements, Unity’s new UI framework, which renders UI for graph-based tools such as Shader Graph, Visual Effect Graph, and Visual Scripting. To help you better organize your complex graphs, Unity has added subgraphs to Visual Effect Graph. You can share, combine, and reuse subgraphs for blocks and operators, and also embed complete VFX within VFX. There is an improvement in the integration between Visual Effect Graph and the High-Definition Render Pipeline (HDRP), which pulls VFX Graph in by default, providing you with additional rendering features. With Shader Graph you can now use Color Modes to highlight nodes on your graph with colors based on various features or select your own colors to improve readability. This is especially useful in large graphs. The team has added swappable Sprites functionality to the 2D Animation tool. With this new feature, you can change a GameObject’s rendered Sprites while reusing the same skeleton rig and animation clips. This lets you quickly create multiple characters using different Sprite Libraries or customize parts of them with Sprite Resolvers. With this release Burst Compiler 1.1 includes several improvements to JIT compilation time and some C# improvements. Additionally, the Visual Studio Code and JetBrains Rider integrations are available as packages. Mobile developers will benefit from improved OpenGL support, as the team has added OpenGL multithreading support (iOS) to improve performance on low-end iOS devices that don’t support Metal. As with all releases, 2019.2 includes a large number of improvements and bug fixes. You can find the full list of features, improvements, and fixes in Unity 2019.2 Release Notes. How to use arrays, lists, and dictionaries in Unity for 3D game development OpenWrt 18.06.4 released with updated Linux kernel, security fixes Curl and the Linux kernel and much more! How to manage complex applications using Kubernetes-based Helm tool [Tutorial]

Last week, the team at PostgreSQL released a progress update for the eagerly awaited PostgreSQL 12. This release  comes with performance improvements and better server configuration, indexes, recovery parameters and much more.   This article was updated 05.14.2019 to correct the fact that this was a progress update for PostgreSQL, not a software release. What’s going to be coming in PostgreSQL 12? Performance In PostgreSQL 12  the Just-in-Time (JIT) compilation will be enabled by default. Memory consumption of COPY and function calls will be reduced and the search performance for multi-byte characters will also be improved. Server configuration Updates to server configuration will add the ability to enable/disable cluster checksums using pg_checksums. It should also reduce the default value of autovacuum_vacuum_cost_delay to 2ms and allows time-based server variables to use micro-seconds. Indexes in PostgreSQL 12  The speed of btree index insertions should be optimized for PostgreSQL. The new code will also improve the space-efficiency of page splits and should further reduce locking overhead, and gives better performance for UPDATEs and DELETEs on indexes with many duplicates. Recovery parameters PostgreSQL 12 should also allow recovery parameters to be changed with reload. These parameters include, archive_cleanup_command, promote_trigger_file, recovery_end_command, and recovery_min_apply_delay. It also allows streaming replication timeout. OID columns The special behavior of OID columns will likely be removed, but columns will still be explicitly specified as type OID. The operations on tables that have columns named OID will need to be adjusted. Data types Data types abstime, reltime, and tinterval look as though they'll be removed from PostgreSQL 12. Geometric functions Geometric functions and operators will be refactored to produce better results than are currently available. The geometric types can be restructured to handle NaN, underflow, overflow and division by zero. To learn more about what's likely to be coming to PostgreSQL 12, check out the official announcement. Building a scalable PostgreSQL solution PostgreSQL security: a quick look at authentication best practices [Tutorial] How to handle backup and recovery with PostgreSQL 11 [Tutorial]  

Last week, Joe Watkins, a PHP developer, shared that PHP 8 will support the Just-in-Time (JIT) compilation. This decision was the result of voting among the PHP core developers for supporting JIT in PHP 8 and also in PHP 7.4 as an experimental feature. If you don’t know what JIT is, it is a compiling strategy in which a program is compiled on the fly into a form that’s usually faster, typically the host CPU’s native instruction set. To do this the JIT compiler has access to dynamic runtime information whereas a standard compiler doesn’t. How PHP programs are compiled? PHP comes with a virtual machine named the Zend VM. The human-readable scripts are compiled into instructions, which are called opcodes that are understandable to the virtual machine. Opcodes are low-level, and hence faster to translate to machine code as compared to the original PHP code. This stage of execution is called compile time. These opcodes are then executed by the Zend VM in the runtime stage. JIT is being implemented as an almost independent part of OPcache, an extension to cache the opcodes so that compilation happens only when it is required. In PHP, JIT will treat the instructions generated for the Zend VM as the intermediate representation. It will then generate an architecture dependent machine code so that the host of your code is no longer the Zend VM, but the CPU directly. Why JIT is introduced in PHP? PHP hits the brick wall Many improvements have been done to PHP since its 7.0 version including optimizations for HashTable, specializations in the Zend VM for certain opcodes, specializations in the compiler for certain sequences, and many more. After so many improvements, now PHP has reached the extent of its ability to be improved any further. PHP for non-Web scenarios Adding support for JIT in PHP will allow its use in scenarios for which it is not even considered today, i.e., in other non-web, CPU-intensive scenarios, where the performance benefits will be very substantial. Faster innovation and more secure implementations With JIT support, the team will be able to develop built-in functions in PHP instead of C without any huge performance penalty. This will make PHP less susceptible to memory management, overflows, and other similar issues associated with C-based development. We can expect the release of PHP 7.4 later this year, which will debut JIT in PHP.  Though there is no official announcement about the release schedule of PHP 8, many are speculating its release in late 2021. Read Joe Watkins’ announcement on his blog. PEAR’s (PHP Extension and Application Repository) web server disabled due to a security breach Symfony leaves PHP-FIG, the framework interoperability group Google App Engine standard environment (beta) now includes PHP 7.2

[box type="note" align="" class="" width=""]This article is an excerpt from a book by Rodolfo Bonnin titled Machine Learning for Developers.[/box] Reinforcement learning is a field that has resurfaced recently, and it has become more popular in the fields of control, finding the solutions to games and situational problems, where a number of steps have to be implemented to solve a problem. A formal definition of reinforcement learning is as follows: "Reinforcement learning is the problem faced by an agent that must learn behavior through trial-and-error interactions with a dynamic environment.” (Kaelbling et al. 1996). In order to have a reference frame for the type of problem we want to solve, we will start by going back to a mathematical concept developed in the 1950s, called the Markov decision process. Markov decision process Before explaining reinforcement learning techniques, we will explain the type of problem we will attack with them. When talking about reinforcement learning, we want to optimize the problem of a Markov decision process. It consists of a mathematical model that aids decision making in situations where the outcomes are in part random, and in part under the control of an agent. The main elements of this model are an Agent, an Environment, and a State, as shown in the following diagram: Simplified scheme of a reinforcement learning process The agent can perform certain actions (such as moving the paddle left or right). These actions can sometimes result in a reward rt, which can be positive or negative (such as an increase or decrease in the score). Actions change the environment and can lead to a new state st+1, where the agent can perform another action at+1. The set of states, actions, and rewards, together with the rules for transitioning from one state to another, make up a Markov decision process. Decision elements To understand the problem, let's situate ourselves in the problem solving environment and look at the main elements: The set of states The action to take is to go from one place to another The reward function is the value represented by the edge The policy is the way to complete the task A discount factor, which determines the importance of future rewards The main difference with traditional forms of supervised and unsupervised learning is the time taken to calculate the reward, which in reinforcement learning is not instantaneous; it comes after a set of steps. Thus, the next state depends on the current state and the decision maker's action, and the state is not dependent on all the previous states (it doesn't have memory), thus it complies with the Markov property. Since this is a Markov decision process, the probability of state st+1 depends only on the current state st and action at: Unrolled reinforcement mechanism The goal of the whole process is to generate a policy P, that maximizes rewards. The training samples are tuples, <s, a, r>.  Optimizing the Markov process Reinforcement learning is an iterative interaction between an agent and the environment. The following occurs at each timestep: The process is in a state and the decision-maker may choose any action that is available in that state The process responds at the next timestep by randomly moving into a new state and giving the decision-maker a corresponding reward The probability that the process moves into its new state is influenced by the chosen action in the form of a state transition function Basic RL techniques: Q-learning One of the most well-known reinforcement learning techniques, and the one we will be implementing in our example, is Q-learning. Q-learning can be used to find an optimal action for any given state in a finite Markov decision process. Q-learning tries to maximize the value of the Q-function that represents the maximum discounted future reward when we perform action a in state s. Once we know the Q-function, the optimal action a in state s is the one with the highest Q- value. We can then define a policy π(s), that gives us the optimal action in any state, expressed as follows: We can define the Q-function for a transition point (st, at, rt, st+1) in terms of the Q-function at the next point (st+1, at+1, rt+1, st+2), similar to what we did with the total discounted future reward. This equation is known as the Bellman equation for Q-learning: In practice, we  can think of the Q-function as a lookup table (called a Q-table) where the states (denoted by s) are rows and the actions (denoted by a) are columns, and the elements (denoted by Q(s, a)) are the rewards that you get if you are in the state given by the row and take the action given by the column. The best action to take at any state is the one with the highest reward: initialize Q-table Q observe initial state s while (! game_finished): select and perform action a get reward r advance to state s' Q(s, a) = Q(s, a) + α(r + γ max_a' Q(s', a') - Q(s, a)) s = s' You will realize that the algorithm is basically doing stochastic gradient descent on the Bellman equation, backpropagating the reward through the state space (or episode) and averaging over many trials (or epochs). Here, α is the learning rate that determines how much of the difference between the previous Q-value and the discounted new maximum Q- value should be incorporated.  We can represent this process with the following flowchart: We have successfully reviewed Q-Learning, one of the most important and innovative architecture of reinforcement learning that have appeared in recent. Every day, such reinforcement models are applied in innovative ways, whether to generate feasible new elements from a selection of previously known classes or even to win against professional players in strategy games. If you enjoyed this excerpt from the book Machine learning for developers, check out the book below.

So far, the articles in this years 12 Days of Christmas series first and second days) have shown some relatively easy fixes for some rather annoying problems. In this article, I will continue that same pattern. In this article, I will share another frustrating problem related to both Kerberos and Managed Service Accounts (MSAs). What is an MSA you might ask? That is a good question. If you are unacquainted with this special type of service account, you have been missing out. An MSA is a special account managed by the domain controller. It is an account that is used to run a service on a single computer (note a Group Managed Service Account, or gMSA, is another type of MSA that can be used on multiple computers). Since the password is managed by the domain and is frequently changed, humans won’t know the password and will not be able to use it to log on to a computer. This type of service account is preferred because of the improved security on the account. That sounds great, right? It is, until you encounter some of the oddball failures and randomness of the service not starting as you might expect. Wayne Sheffield addressed some of the problems relating to the service starting in his article here. As luck would have it, there is a similar problem that needs just a little more action on our part in order to nudge the service to be able to start consistently. Fix SQL Service not Starting after Reboot when using an MSA As a general rule of thumb, I recommend adding the service dependencies to your SQL Server Service. These service dependencies are (as mentioned in Wayne’s article): Netlogon, KEYISO, and W32Time. Additionally, I recommend that you double check the dependency for the SQL Server Agent to ensure it is dependent upon SQL Server Service. Most of the time, the agent service will already have the SQL Server Service dependency and the SQL Server Service will have the KEYISO dependency. However, that is not a foregone conclusion. Occasionally, those dependencies are missing. Netlogon and W32Time are almost always missing so you will need to add those dependencies most of the time. I won’t go into further detail about setting these dependencies today. Wayne showed how to use regedit to set the dependencies. That said, these dependencies should be set prior to setting the “fix” I will mention shortly in this article. In addition, stay tuned because I will be showing in the near future how to set those dependencies via PowerShell. Occasional Errors after Reboot The problem of SQL Server Service not starting after a reboot when using an MSA is ridiculously frustrating. Why? Well, you get very little information as to what the problem is. The event viewer just shows (most of the time) that the service stopped and doesn’t give an error code. In fact it just shows it as an informational event. Every once in a while, you may get a gold nugget of a clue with a message similar to the following: NETLOGON error: This computer was not able to set up a secure session with a domain controller in domain TurdWilligers due to the following: There are currently no logon servers available to service the logon request. This may lead to authentication problems. Make sure that this computer is connected to the network. If the problem persists, please contact your domain administrator. ADDITIONAL INFO If this computer is a domain controller for the specified domain, it sets up the secure session to the primary domain controller emulator in the specified domain. Otherwise, this computer sets up the secure session to any domain controller in the specified domain. Despite the service dependency on Netlogon, you can still see SQL Server try to start before the Netlogon service is fully functional. This is disastrous, right? Not exactly. Despite the flawed attempt to start a bit quick there, we still have our methods to avoid this problem in the future. You will laugh at how ridiculously easy this fix is. Are you ready for it? And now the Fix The fix is (after setting the service dependencies mentioned previously) to set the SQL Server Service to “Automatic (Delayed Start)” instead of “Automatic” as shown in the following image. All that needs to be done is this quick and simple change in the services control panel (not SQL Server Configuration Manager). Once applied, reboot the server to validate that it worked. TADA – problem solved. Now, take this fix and make your environment consistent. Or not. The choice is yours. Personally, I prefer to have my environment as consistent as absolutely possible. There are times when that is not plausible and it is these times that it is very highly recommended to DOCUMENT what is different and the reasons as to why it is different. This makes your job significantly easier and gives you back a significant amount of time that you will grow to appreciate more and more over time. Put a bow on it Running into startup issues with your services could be a nightmare on some days. Couple those issues with an MSA and you may be looking at a bald spot on your head before long. Never fear, this article demonstrates that startup issues involving MSAs can be an extremely simple condition to resolve. The most benign of changes affecting your service startup can be the difference between SQL Server starting normally after an unexpected reboot or SQL Server being down until somebody “notices” the problem. As mentioned previously in this article, please stay tuned for future tips on how to simplify adding the service dependencies. It should prove a useful tool in your arsenal, helping you to become more efficient and elite. Enjoyed reading some of these security related articles? Security is a fantastic topic and there is plenty to learn in regards to security. Try this article for more related to Kerberos issues. Or maybe you want something a little more in your face about security? Try this back to basics article about public role permissions. This is the third article in the 2020 “12 Days of Christmas” series. For the full list of articles, please visit this page. The post Here is an Easy Fix for SQL Service Startup Issues When Using an MSA first appeared on SQL RNNR. Related Posts: CRM Data Source Connection Error January 23, 2020 SHUTDOWN SQL Server December 3, 2018 Single User Mode - Back to Basics May 31, 2018 The Gift of the SPN December 10, 2019 How To Resolve User Error in Kerberos Configuration Manager December 26, 2020 The post Here is an Easy Fix for SQL Service Startup Issues When Using an MSA appeared first on SQLServerCentral.