AlphaFold

What is AlphaFold?

AlphaFold is a Artificial intelligencewhich is able to predict the three-dimensional protein structure using only the amino acid sequence of the protein. In 2020, AlphaFold2 has achieved the accuracy of experimental methods, solving a 50-year-old problem in biology: the problem of protein folding.

Since 2021, AlphaFold DB has been available as Database of protein structures is freely available to scientists around the world. In addition, a free version of AlphaFold2 as open source software at Github be used.

The solution to the protein folding problem

AlphaFold is another success story from Google DeepMind (AlphaGo, AlphaZero and Ithaca). Shortly after AlphaGo clearly defeated the famous professional player Lee Sedol in GoIn 2016, a new DeepMind team was put together. Its goal was to solve the problem of protein structure prediction.

2018 marks the team's first success: AlphaFold takes first place in the 13th CASP competition.

The CASP; Critical Assessment of Techniques for Protein Structure Prediction (in German: kritische Bewertung von Techniken zur Vorhersage von Proteinstrukturen) is an association of scientists who have been researching the problem of protein folding since 1994. Every two years, a competition takes place in which research teams are given a selection of amino acid sequences for proteins. Their exact three-dimensional shape is already known, but not publicly available. The teams give their best predictions to see how close they are to the actual structures.

The AlphaFold's outstanding results in this competition are published in the journal Nature and the DeepMind team continues to expand. In 2020, the breakthrough is made: AlphaFold2 wins the 14th CASP competition by a large margin and is recognised by the CASP organisers as the solution to the 50-year-old "protein folding problem". AlphaFold2 can predict protein structures to atomic accuracy with a mean error (RMSD_95) of less than 1 angstrom, making it three times more accurate than the next best system and comparable to experimental methods. Among experts, the solution of the protein folding problem is dubbed one of the most important achievements since the mapping of the human genome.

How does AlphaFold work?

In principle, AlphaFold uses Neural networks which through Deep Learning be trained. After the very good results with CASP13, the programme is being further developed. Its methods and code are published in the scientific journal Nature. This results in Open source implementations from the community. DeepMind itself has added new deep learning architectures and further developed the methods.

Crucial to these methods are the fields of biology, especially in the area of protein folding, as well as physics and machine learning.

To understand the physical interactions within proteins, it is important to understand how a folded protein is constructed. It can be seen as a "spatial graph" in which residues are the nodes and edges connect the residues in close proximity. For AlphaFold2, which wins CASP14, an attention-based neural network system is used. With this, it continuously tries to interpret the structure of the protein while thinking about the diagram it is assembling. Multiple sequence alignment (MSA) and a representation of amino acid residue pairs are used to improve this diagram.

By constantly repeating this process, AlphaFold2 achieves strong predictions of protein structure. By using its own confidence measure, AlphaFold2 is also able to determine which parts of its predicted protein structure can be classified as reliable.

AlphaFold Protein Structure Database

In close cooperation with the European Bioinformatics Institute at the European Molecular Biology Laboratory (EMBL-EBI) DeepMind 2021 launches the AlphaFold Protein Structure Database. Thus the Scientific community free and open access to the human proteome (the totality of all proteins in the human body) together with 20 other model organisms, including mice. The database thus comprises a total of over 350,000 structures. At the beginning of 2022, DeepMind will add another 27 proteomes (corresponding to over 190,000 proteins) to the database.

To date, over 300,000 researchers worldwide have made use of the database. This represents AlphaFold one of AI's most significant contributions to advancing scientific knowledge dar.

AlphaStar

What is AlphaStar?

AlphaStar is the first Artificial intelligencewhich is the real-time strategy game StarCraft 2 mastered at an outstanding level. For this it uses machine learning and deep neural networks. According to AlphaGo, AlphaZero and AlphaFold AlphaStar is another Google DeepMind success.

What is StarCraft 2?

StarCraft 2 is a real-time strategy game by Blizzard Entertainment and is considered one of the most complex and popular of all e-sports games. There are a few things to keep in mind when playing:

Activision Blizzard

1. Micro-Managementalso called micro: The player must permanently direct dozens of game pieces on the playing field at the same time and give orders.

2. Macro managementcalled Macro for short: The player must develop a suitable strategy for constructing buildings, researching upgrades and building new game pieces.

3. Incomplete information: Due to the "fog of war", the enemy's actions are usually not visible. Your own units make the fog disappear within a small radius. Therefore, you have to send them into the enemy's areas to get a glimpse of their tactics. This action is called "scouting".

4. the Rock-paper-scissors principle: every attack can be answered with a suitable defence. This in turn gives an advantage to the counter-attack.

5. everything happens in Real-time: Unlike in chess or go, both players must plan and execute their actions simultaneously and without pause.

6. Three different breedsThere are three different alien races to choose from at the beginning of the game: Protoss,Terran and Zerg. All have their own characteristics and allow for different strategies.

Function and playing strength of AlphaStar

For artificial intelligences, this Complexity a major challenge s. AlphaStar has to choose from up to 1026 possible actions each time, with many more moves whose effects can only be seen after a long period of play. To develop successful strategies for victory, AlphaStar had to learn and further research human strategies.

To do this, it uses various general machine learning techniques, including self-play with reinforcement learning and imitation learning, combined with Neural networks. This makes it AlphaStar to earn a Grandmaster level for all three alien races and play better than 99.8 % of StarCraft players on Battle.net.

AlphaStar's strong results provide evidence of how general learning techniques can strengthen AI systems so that they can function in dynamic and complex environments. These results are expected to be applicable in the real world in the future, making AIs safer and more robust.

Online resources

Github

Since 2017, there has been the StarCraft API PySC2 as open source at Github. It is the result of a cooperation between DeepMind and Blizzard Entertainment. Through the API, computers can receive information directly using feature maps without having to evaluate the game graphics. For example, a feature map can consist of a matrix whose cells indicate the type of a character (for example, 0 for "no character"). This data is much easier for computers to process. Various artificial intelligences have since used the PySC2 API and compete against each other as StarCraft bots.

Nature

DeepMind has published its research results on AlphaStar in the scientific journal Nature. In it you will find Information on the techniques used by AlphaStar and their benefits.

Autonomous driving

What is autonomous driving?

Autonomous driving basically describes controlling a vehicle or system without a driver. The term is by no means limited to automobiles, but can also be applied to Robot or transport systems, although the term is widely used in the automotive sector.

5 levels of autonomous driving

The degree of automation is classified from 0 to 5 using an internationally standardised level. Level 0 describes the self-driver who drives without the help of driver assistance systems. Levels 1 to 5 of autonomous driving are described below:

  • At Level 1 (Assisted driving), the system takes over certain tasks of driving the vehicle, whereby the driver must be able to take full control of the vehicle at all times. An example of this is the use of cruise control, lane departure warning or distance assistance.
  • Level 2 (Semi-automated driving) enables the vehicle to temporarily take over driving tasks from the driver. Examples of this are the overtaking assistant or automatic parking, where the driver is allowed to take his hands off the steering wheel, but must still be able to take control of the vehicle at any time.
  • Autonomous driving Level 3 (Highly automated driving) goes a step further in terms of the driving tasks that the system can perform without human intervention. The driver is allowed to temporarily turn away from the traffic and, for example, read the newspaper or take care of passengers in the back seat. However, the driver must be able to resume driving the vehicle after a sufficient time has elapsed. The vehicle manufacturers define the conditions for the operation of highly automated driving. Motorways are particularly suitable for this, as there is no oncoming traffic here and the road markings are usually clearly visible.
  • In Level 4 (Fully automated driving), the level of automation of the vehicle increases by further components. Autonomous driving level 4 allows the driver to completely hand over control of the vehicle. Even the presence of a passenger during the journey is not necessary, so that independent parking procedures or motorway journeys of the vehicle are also possible. If the system reaches its limits, it is able to reach a safe state such as a parking space or the edge of the road. At this level, however, the driver still has the option of taking over all driving tasks himself.
  • In the last Level 5 (Autonomous driving) of automation, the driving operation is only controlled by the system, whereby all vehicle occupants permanently become passengers and no longer take on driving tasks, as the system solves all situations independently.

Where do we stand today?

While level 1 and 2 autonomous driving functions are already widespread in most new cars, offers Mercedes-Benz is the first manufacturer to have an approved Level 3 system for the S-Class and EQS for sales in Germany to. The Drive Pilot can be activated and used in defined situations on motorways up to a speed of 60 km/h.

Other German car manufacturers or suppliers such as Audi, BMW, VW or Bosch are also researching and in some cases developing in cooperation in the field of autonomous driving. Some technology companies such as Google (with its subsidiary Waymo) and Apple are conducting intensive research in this area. Tesla currently delivers their vehicles with a hardware version that would allow autonomous driving level 5, but this is not activated due to a lack of approval.

In the field of transport logistics, such as at MAN or DB Schenker, research is being conducted on driverless transport vehicles.

In addition to the technical requirements for the individual stages of autonomous driving, the legal component also plays a decisive role, whereby the VDA (German Association of the Automotive Industry) supports the creation of a regulatory framework in order to act as a technology driver in Germany.

Content of the VDA flagship initiative

The VDA flagship initiative autonomous and connected driving (AVF) has set itself the goal of strengthening Germany as a location for automated and connected driving through cooperation and increased transparency in pre-competitive basic research by building up broad-based expertise. The flagship initiative is intended to coordinate development needs and funding priorities, leverage synergies between projects and avoid redundancies and misinvestments.

AlphaZero

What is AlphaZero?

AlphaZero is a self-learning computer programme from Google DeepMind that learns the board games Go, Chess and Shogi using its algorithm. For this it combines machine learning with neural networks. In order to learn one of the board games mentioned, the Artificial intelligence only the rules of the game, playing conditions and intensive playing against itself. It is not necessary to be taught by humans.

AlphaZero is based on the approach of AlphaGo Zerowhich was also developed by DeepMind.

How powerful is the programme?

Since AlphaZero is self-learning, its performance increases with every match played. It only knows the respective rules of the game and starts playing against itself with random moves. It evaluates its moves according to the result and thus learns which strategies work best.

After just 4 hours, AlphaZero was able to play chess at a superhuman level, surpassing the playing strength of world chess champion Magnus Carlsen with an Elo rating of over 2800. After 9 hours of training, it surpassed even the best chess programme up to then, Stockfish 8, with a calculated Elo rating of 3300.

After 34 hours AlphaZero has mastered the board games Chess, Shogi and Go and challenged all the leading programmes up to that point. Besides winning against Stockfisch 8 in chess, Elmo did it in shogi and AlphaGo defeated in Go.

It should be noted, however, that for the immense number of games against itself, very much computing power needed is. To make this possible, a large amount of TPUs (tensor processors to speed up machine learning applications) from Google were used. This allowed AlphaZero to complete approximately 44 million chess games in 9 hours, 24 million shogi games in 12 hours and 21 million Go games in 34 hours, putting AlphaZero's extremely rapid learning success into perspective.

Important games

AlphaZero and Stockfish

Stockfish is a chess programme, which, unlike AlphaZero, trained by humans became and as Open Source is available free of charge. Stockfish's algorithm uses a huge repertoire of moves, which it has learned by playing against human players. AlphaZero lacks this human influence, which is why the game is considered unconventional.

In 2017, Stockfish 8 was the best chess program and was clearly beaten by AlphaZero. This was to prove that a self-learning algorithm is superior to a human-trained algorithm.

2021 Stockfish 14 (the latest version) is again the number 1 chess programThis was made possible, among other things, by the cooperation with Leela Chess Zero and the use of neural networks.

AlphaZero and Leela Chess Zero

Leela Chess Zero (Lc0 for short) is a chess program, which can be used free of charge as open source, just like Stockfish. However, it was modelled on AlphaZero in its mode of operation. Leela Chess Zero is based on Leela Zero Go, which in turn is modelled on Alpha Go Zero. Accordingly, Lc0 uses its artificial intelligence combined with machine learning and neural networks to learn chess. Indeed, one of the main purposes for developing Leela Chess Zero is to validate the methodology of AlphaZero.

There is no official match of AlphaZero against Leela Chess Zero. For this, Lc0 regularly competes against Stockfish and is now at a similar level. In 2021, Lc0 was in 2nd place in the ranking of the best chess programs, behind Stockfish 14.

AIOps

What is AIOps?

On the one hand, AIOps stands for Artificial Intelligence for IT Operations. - Artificial intelligence for IT operations". This is a subcategory of the Artificial intelligencewhich was created by the US research and consulting company Gartner in 2016.

On the other hand, the Acronym supplemented by the meaning "Algorithmic IT Operation" - which means the use of Machine learning for analysis purposes.

The need for this particular field is explained by the modern IT landscape. Static and predictable systems must give way to software environments that change and reconfigure during operation. In addition, data will in future be less often stored in data centres or Clouds generated, but by billions of networked IoT-devices. Against this background, the installation of AIOps within companies with a broad digital infrastructure is inevitable in order to remain competitive.

How is AIOps different from DevOps?

DevOps describes the working culture that is necessary to develop software systems efficiently and effectively. This includes methods such as automated development processes, agile teamwork or decoupled processes that communicate with each other via programming interfaces.

AIOps also refers to the culture as a prerequisite for the targeted development of software. Topics such as Scrum, development pipelines or automation (they are otherwise associated with DevOps) also play a major role here. However, the Focus on the use of Big Data and machine learning - i.e. working with artificial intelligence - in order to, among other things, achieve the following Analyse and optimise IT operational processes:

  • In event correlation, so-called events are monitored. These are logins and events that are carried out during a computer session. The aim is to identify operational errors and uncover causes.
  • Anomaly detection is used as a further measure of danger prevention within the framework of AIOps. While event correlation concentrates on the detection of already known dangers, anomaly detection detects a communication pattern that deviates from the usual behaviour of the system. Cyberattacks, for example, can be unmasked with this.
  • As the name suggests, causality identification strives to show and explain connections. Depending on the technical approach, this idea is implemented individually by the well-known AIOps platforms.

What platforms and tools are there?

AIOps platforms:

  • The computer programme IBM Watson Understands natural language and Responds to questions. With the Cloud Pak platform, it also has an efficient AIOps solution to collect data from various sources and centralise the information derived from it in one point - a solid and proven way to automate an IT operation.
  • The provider Splunk originally focused on logging, monitoring and reporting with the platform of the same name.. With the additional AIOps division, customers are supported in centralising and simplifying analysis.
  • Also the provider Aruba combines Big Data and machine learning to automate IT operations processes.

AIOps tools:

  • The Market research company Gartner predicted the direction in which the IT sector was heading early on. From this, one can deduce the expertise they have in this area. Based on research findings, Gartner helps clients make informed decisions and develops tools for a wide range of business areas and market analyses. One tool in particular is relevant for the AIOps sector: the Gartner Hype Cycle.