Artificial Intelligence in Supply Chain Management. Opportunities and risks of digitalisation

Masterarbeit , 2020 , 134 Seiten

BWL - Beschaffung, Produktion, Logistik

Industrialisation has greatly changed people's lives since the eighteenth century. At that time, the focus of activities shifted from the physical to the cognitive. We are now in the fourth industrial revolution. Through the development of ever better artificial systems, more and more cognitive activities are being carried out by machines.

But what exactly does artificial intelligence mean? What changes has it already created in supply chain management and what is still possible in the future? What are the risks associated with advancing digitalisation?

In his publication, Johannes Hangl creates a comprehensive picture of the status of current and future developments as well as trends in the use of artificial intelligence. He shows what possibilities, effects, opportunities and risks it has for supply chain management.

From the contents:
- AI;
- Internet of Things;
- Logistics;
- Cyber-Physical Systems

Table of contents

List of abbreviations

1 Introduction
1.1 Problem statement
1.2 Objective
1.3 Working methodology
1.4 Course of work

2 Theoretical foundations
2.1 Supply Chain Management
2.2 artificial intelligence

3 Areas of application and current research on artificial intelligence in the supply chain
3.1 Process Overview Supply Chain
3.2 Procurement
3.3 Production
3.4 Intralogistics and warehousing
3.5 Distribution
3.6 Compliance, Customs, Import and Export
3.7 Disposal and recycling

4 Opportunities and risks of artificial intelligence in the supply chain
4.1 Odds
4.2 Risks
4.3 Summary of the use of artificial intelligence in the supply chain

5 Conclusion

Bibliography

List of abbreviations

Abbildung in dieser Leseprobe nicht enthalten

1 Introduction

1.1 Problem statement

With the beginning of industrialization, people's lives have changed a lot and brought prosperity and a better life for many. With the beginning of the first industrial revolution at the end of the 18th century (first mechanical loom in 1784), work has changed from an often purely physical activity to a more cognitive activity. This was made possible by the development of mechanical looms and further advances in the field of hydropower and steam power.¹

Automation was strengthened by the second industrial revolution at the beginning of the 20th century with the introduction of the first assembly lines (first assembly line in 1870) and the introduction of mass production based on the division of labor using electrical energy.²

The third industrial revolution, which was initiated in 1969 with the first programmable logic controller (PLC), has contributed to the further automation of production through the use of electronics and information technology.³

The fourth industrial revolution in which society finds itself today is based on cyber-physical systems (CPS).⁴ CPS is a network of computer, software components with mechanical and electronic parts that communicate via a data infrastructure, such as the Internet.⁵

Due to the development of ever better artificial systems, cognitive activities are carried out by machines and this is now often more optimal and reliable than a human could ever do.⁶

Stephan Hawking made headlines when he explained in an interview that "the development of a comprehensive (strong) artificial intelligence could mean the end of humanity".⁷

Numerous institutions, companies and associations are concerned with the further development of artificial intelligence. Among the great successes of the development are, for example, the computer Deep Blue, which won in 1997 against the chess world champion Garry Kasparov, but also Watson of IBM, which in 2011 the quiz show ' Jeopardy! ' has won.⁸

In recent years, this development has accelerated and already today the use of artificial intelligence in contract analysis is already much more effective than top lawyers.⁹

Due to the advancing globalization and the worldwide collaboration between companies at all levels, the exchange of goods, services, data and information is becoming increasingly important. The reason for this is also that artificial intelligence requires considerable evaluable amounts of data in order to learn from them. The areas of logistics, supply chain management and information technology are becoming increasingly important and are regarded as an elementary competitive factor in a networked world.¹⁰

According to Kuhn and Hellingrath, supply chain management can be defined as integrated process-oriented planning and control of the flows of goods, information and money along the entire value chain from the customer to the supplier with the goals

- Improvement of customer orientation,
- Synchronization of the supply with the demand,
- Flexibilization and needs-based production,
- Reduction of stocks along the value chain,
- Efficient material management including disposal and recycling.¹¹

Artificial intelligence can help meet these goals. However, the possibilities, opportunities and risks are not clear. These previously outlined developments lead to the question:

What possibilities does artificial intelligence offer in the supply chain? What opportunities and risks are associated with this?

1.2 Objective

The aim of the work is to provide a comprehensive picture of the state of current and future development as well as trends in the use of artificial intelligence. Furthermore, it will be shown what possibilities, effects, opportunities and risks artificial intelligence has for supply chain management.

In order to meet the aforementioned objective, the following sub-objectives are defined:

- Creating a fundamental understanding of supply chain management
- Creating a fundamental understanding of artificial intelligence
- Explanation of developments over time
- Current use of artificial intelligence along the supply chain
- Explaining the opportunities and risks of artificial intelligence

1.3 Working methodology

In order to present a comprehensive picture of the current state of development, future developments and trends of artificial intelligence and to show what possibilities, effects, opportunities and risks this has for supply chain management, both a quantitative and a qualitative study was carried out.

To answer the research question, both the qualitative content analysis was carried out and a literature search was carried out. To this end, publications of the last fifty years on the keywords artificial intelligence, artificial intelligence, supply chain and logistics were examined as part of the literature search. Subsequently, publications in journals, blogs and newsletters were examined in order to be able to analyze and classify current developments, opportunities, risks and future developments.

In order to get an overview of the literature on artificial intelligence and supply chain management, relevant books, journal articles, blogs and expert interviews were determined using the THM library portal, Google Scholar and the Google search engine.

With the above-mentioned tools, 678 relevant sources of information were identified by entering search terms. The 678 sources include books, scientific journals, journal articles, newspaper articles, magazines, university publications, laws and regulations, interviews and Internet documents.

The literature was analyzed, coded and evaluated on the basis of certain search terms. The sources were provided with about 85,000 codes using software and classified into 53 different categories. The codes were then sorted according to relevance, timeliness and relation to the specific question and then analyzed.

For example, all texts in which the keywords opportunities in connection with artificial intelligence appear in a paragraph were provided with the code chances. These coded passages were subsequently ordered and analyzed according to their relevance and topicality. Relevant correlations were referenced and presented in the work.

The literature used in this work represents the distillate of literature analysis; the literature was selected on the basis of its topicality and relevance to the research topic

1.4 Course of work

Chapter 1 defines the problem, objective and working methodology of this work.

Chapter 2.1 presents the basic knowledge of supply chain management, its tasks and objectives as well as the typical problems. Subsequently, Chapter 2.2 explains artificial intelligence and its sub-areas. In this context, it is also important to explain technical terms such as big data, machine learning, robotics, neural networks, algorithms and programs, cognitive science and Industry 4.0 as well as the Internet of Things (IoT).

Chapter 3 presents both internal and external areas of application, but it will also discuss other different application areas of artificial intelligence both in the Business to Business (B2B) segment and Business to Customer (B2C) segment. This will be explained using current examples.

Chapter 4 sets out the opportunities associated with artificial intelligence, but also the risks. Furthermore, it is examined how companies can cope with the transformation to the digital factory (Smart Factory) and what risks are associated with it.

This work is concluded in Chapter 5 with a conclusion in which the core topics are briefly summarized.

2 Theoretical foundations

Chapter 2 presents the theoretical foundations of supply chain management and artificial intelligence. Chapter 2.1 deals with the definition of supply chain management and the explanation of tasks and objectives; In addition, typical problems are addressed and an explanation of the Supply Chain Operations Reference Model (SCOR Model) is given.

Chapter 2.2 deals with the definition, tasks and objectives of artificial intelligence. Furthermore, the technical terms and the basics of artificial intelligence are explained in more detail.

2.1 Supply Chain Management

2.1.1 Definition of Supply Chain Management

A supply chain can be understood as a supply chain that ranges from the raw material supplier (point of source) to the end user (point of consumption). A supply chain or supply chain includes all material, information and cash flows along the entire value chain – internally and externally.

Abbildung in dieser Leseprobe nicht enthalten

Figure Schematic representation of a supply chain from the point of view of a production company

By source cf. Kuhn/Hellingrath (2002) and cf. Weetman (2016)

In principle, value creation measures the self-created services of a company minus previous and/or third-party services. Until now, the individual divisions, such as sales and purchasing, usually functioned independently of each other.¹²

In the supply chain, all internal as well as external partners (raw material suppliers, suppliers, distributors and dealers) are involved in the value creation process.¹³

In the context of an initial conceptual clarification of supply chain management, the opinions of various authors and schools of thought sometimes differ significantly. Regardless of the different definitions, it is generally accepted that supply chain management is based on Porter's value chain.¹⁴ However, in supply chain management, the interfaces between the interacting companies are also included.

For this work, the following definition of supply chain management is used:

According to Kuhn and Hellingrath, supply chain management is the integrated process-oriented planning and control of goods, information and cash flows along the entire value chain from the end customer to the raw material supplier with the goals

- Improvement of customer orientation
- Synchronization of the supply with the demand
- Flexibilization and needs-based production
- Reduction of inventories along the value chain
- Efficient material management including disposal and recycling.¹⁵

Supply chain management is a process science that should be process-oriented and integrated. Integrated in this context means that people no longer think separately according to functions and in rigid hierarchies. The aim is to create interdisciplinary teams that work together in partnership both internally and externally in order to optimize value creation along the entire process.¹⁶

For example, when developing a new product, employees from other areas are integrated in addition to the developers. It makes a good case to involve employees from purchasing and logistics already in product development, in addition, if necessary, employees of the supplier or the up-supplier. Savings potential can already be exploited during the product development process.

Merten defines the following supply chain management philosophy:

" The optimization of the overall system is better than the optimization of subsystems. "¹⁷

The process within supply chain management can be displayed in an Order-to-Payment-S process. This process maps the sequence of activities that are necessary to process a sales order.¹⁸ The process can be divided into three areas.

Abbildung in dieser Leseprobe nicht enthalten

Figure Order-to-Payment S. process

By source. cf. Werner (2013), p. 9

Area 1: Upstream (from right to left): The Client places an order with the Company. (Pull orientation). The dispatchers use delivery schedules to control the orders in order to derive the parts to be manufactured. The information obtained is made available to the purchasing department in order to ensure the replenishment of goods.

Area 2: Downstream (from left to right): Here, the fulfillment of the customer order is the focus of the actions. The goods are accepted at the goods receipt, processed, converted or processed after storage and then delivered to the customer as finished goods.

Area 3: Upstream (from right to left): Finally, the goods are paid for by the customer. Disposal or recycling also runs in this direction. Opportunity costs should be avoided.¹⁹

2.1.2 Tasks and objectives of supply chain management

The central task of supply chain management is the efficient and effective alignment of all activities and processes along the entire value chain with the customer. In order to define the efficient and effective orientation towards the customer more precisely, one can orient oneself on the 7R or 8R+ of logistics:

Abbildung in dieser Leseprobe nicht enthalten

In today's time of digitization,²⁰ especially due to the Internet of Things, the right information becomes an important factor, as well as the right packaging, which is equipped with RFID or other sensors. Therefore, the classic 7Rs are extended to 8R+. The + means that more components such as the right packaging could be added.

In order to fulfill the above tasks and satisfy the customer, supply chain management aims to optimize the effectiveness and efficiency of the company's activities and to harmonize the competitive factors of cost, time, quality and flexibility.²¹

Furthermore, according to Lenzen, cross-company success potentials are to be tapped. This is only possible if the companies cooperate in a supply chain and common goals are set.²²

Supply chain management does not clearly distinguish itself from logistics, but either from other areas such as purchasing, procurement and materials management. The boundaries are fluid. In logistics, the main focus is on the physical flow of materials within and between companies, organizations and customers and thus on bridging space and time.²³

Both concepts are based on flow and process orientation, but supply chain management deals with the holistic view of the logistics chain, including all money and information flows across one's own company boundaries.²⁴

2.1.3 Typical problems of supply chain management

According to Werner, supply chain management emerged as a result of the following phenomena:

- Bullwhip effect
- Total Cost of Ownership
- transaction cost
- globalization²⁵

2.1.3.1 Bullwhip effect

The bullwhip effect, or in German the whip impact effect, goes back to Forrester's investigations from 1958. Forrester found that when demand increases, players in a value chain (raw material supplier, manufacturer, distributor, distributor, and customer) overreact.²⁶ It has been shown that even an unplanned increase in demand of 10% causes the manufacturer to increase production by up to 40%. This can also be seen in Figure 3.

Abbildung in dieser Leseprobe nicht enthalten

Figure Bullwhip effect

By source: Knolmayer/Mertens/Zeier (2000)

For example, the end consumer has a short-term increase in demand of 10% due to an offer or a bottleneck situation (discount campaign, bottleneck, hamster purchases) by the retailer. The retailer orders an increased quantity from the up-supplier. Across the stages, the demand forecast increases because the individual market partners can only see the needs of the respective upstream stage. In addition to the lack of transparency of demand, the reasons for this are the distortion of information, price changes and changes in inventory levels.²⁷

To mitigate or combat the bullwhip effect, there are certain tools. Above all, an improved exchange of information along the entire supply chain is needed in order to be able to determine the actual demand of the end customer.²⁸

According to Schulte, Figure 4 explains the causes of the bullwhip effect and sets out possible solutions:

Abbildung in dieser Leseprobe nicht enthalten

Figure Causes and possible solutions of the bullwhip effect

Source:Schulte (2017), p. 20

According to Schulte, forecast-based production plans, accumulation of orders, price promotions and product allocation are the four causes of the bullwhip effect.

The bullwhip effect in the area of forecast-based production plans can be solved by transparency from the point of sale to the raw material supplier. The IT networking of the individual stages with each other is particularly suitable for this purpose.

In order to avoid the accumulation of orders, an even order distribution should be sought. By communicating along the entire supply chain network, orders can be better consolidated and planned.

In order to avoid fluctuations in demand regarding price promotions, price promotions should be communicated and planned in advance. Any fluctuations should be taken into account. By using intelligent allocation mechanisms, bottlenecks and phantom orders can be avoided.

2.1.3.2 Total Cost of Ownership, Transaction Costs and Globalization

Another motive for the emergence of supply chains lies in the increasingly complex procurement, production and distribution processes. Due to the growing global network (globalization), goods can be procured worldwide at any time and anywhere in the world. This makes both the procurement process immensely complex and, on the customer side, the delivery to customers worldwide. In addition to factors such as faster delivery, good quality and other points of the 8 R+, the costs also play a role.

Through billing procedures such as the Total Cost of Ownership (TCO), not only the acquisition costs, but all cost aspects are included in the calculation.²⁹

For example, in a procurement process of merchandise, freight costs, customs costs, storage costs and capital costs are included in the calculation in addition to the purchase price. Factors such as the delivery time also play a role here. These have a not insignificant share of the safety stock.

In addition to economic liberalization, cheap and fast transport routes as well as the expansion of communication options are constantly presenting companies with new challenges. However, this also increases competition between companies. If the customer can shop anywhere at any time, the company must be able to react quickly, flexibly and efficiently.³⁰

2.1.4 The Supply Chain Operations Reference Model (SCOR Model)

In order to be able to optimize the business processes across several stages and companies, a general standard process or a standard sequence is required. This general standard is provided by the Supply Chain Operations Reference Model (SCOR), which was developed by the Supply Chain Council.³¹

Abbildung in dieser Leseprobe nicht enthalten

Figure SCOR Model

Source:BearingPoint (2002)

The SCOR model describes on a general level the business processes that take place in a supply chain. It contains standardized descriptions of the processes for the design, planning, control and execution of the material and information flow in cross-company value creation. Standardized key figures and best practices are assigned to all processes.³²

The SCOR model is hierarchical and industry-independent. At the highest level (level 1), the SCOR model shows the four main processes. The scheme and a brief description of the individual levels are shown in the Figure 5 depicted.

2.1.4.1 Level 1

According to Ziegenbein, the five main processes are planning, procuring, manufacturing, delivering and return. Planning includes all processes and activities with which the expected resources in procurement, production and sales are aligned with the expected demand. Procurement, manufacture and delivery includes all processes and activities of the procurement of services and goods, their treatment and processing as well as the delivery to the customer. Return includes all activities and processes for the return of goods.³³

2.1.4.2 Level 2

At the configuration level, the main processes are subdivided according to whether the products are procured, produced or sold in stock, on sales order or according to customer specifications.³⁴

2.1.4.3 Levels 3 and 4

At the other levels, where level 4 is not included in the SCOR model, level 2 is presented in even more detail and also applied to specific industries.

"All processes in the SCOR model contain a defined process description with input and output variables, key performance indicators and best practices."³⁵

The great advantage of this model is that the same processes and tools are applied to all collaborating companies in a common supply chain and thus homogenization is possible. This is particularly useful when IT systems network the individual companies with each other.

2.2 artificial intelligence

2.2.1 Definition of artificial intelligence

Artificial intelligence or AI can be interpreted and interpreted in a variety of ways. John McCarthy, one of the pioneers of artificial intelligence, defined artificial intelligence as the science of constructing intelligent machines.³⁶

From today's point of view, this definition is not sufficient to do justice to the subject area. A generally accepted definition is not yet available. For this work, the following definition of artificial intelligence by Lackes and Siepermann is used, which states that artificial intelligence deals with methods that allow a computer to solve such tasks that, when solved by humans, require intelligence.³⁷

2.2.2 Tasks and objectives

In order to determine whether a computer or machine has artificial intelligence, Alan Turing developed a test as early as the beginning of 1950. the Figure 6 shows the schematic representation of this test.

Abbildung in dieser Leseprobe nicht enthalten

Figure Turing Test

By source: Banfi (2018)

The questioner (A) should determine by targeted questions or by communication, which of the two communication partners is the human (B) and who the computer (C) is. The difficulty is that both communication partners B and C are not visible and the questioner communicates with the two by means of written communication. If the questioner cannot make a clear distinction or identify computer C as a human, the computer has a certain intelligence.

Since the beginning of 1990, the so-called Turing Prize has been announced, in which the system that passes the Turing test wins a prize. In 2014, this test was passed for the first time. Artificial intelligence (Eugene Goostman) convinced 33% of examiners that they were communicating with a 13-year-old child.³⁸

In May 2018, Google went one step further and was able to introduce the 'Duplex' system, a program that uses natural language to book a hairdressing appointment, for example. It is no longer recognizable that it is a computer.³⁹ Meanwhile, there are other tests such as the Lovelace test or the Metzinger test.

Artificial intelligence can be divided into two areas, the strong and the weak artificial intelligence.⁴⁰

Abbildung in dieser Leseprobe nicht enthalten

Figure Weak and strong artificial intelligence

By source: cf. Jaekel (2017), p. 16

2.2.2.1 Weak artificial intelligence

Weak AI is a system that focuses on solving specific application problems. Weak artificial intelligence thus deals with a clearly defined task in a clearly defined area. The approach to the problem always remains the same and follows the same flow pattern.⁴¹

Weak AI systems are already in use in numerous areas today. Some examples are navigation systems, image and speech recognition, automatic translation, expert systems, chatbots and many other applications.

2.2.2.2 Powerful artificial intelligence

By strong artificial intelligence (also referred to as superintelligence, or strong AI or general AI), Buxmann and Schmidt understand systems that have the same or higher intellectual abilities than a human.⁴²

According to the University of Oldenburg, the characteristics of a strong artificial intelligence are:

1. reasoning
2. Making decisions in case of uncertainty
3. plan
4. learn
5. Communication in natural language
6. Use all of these skills to achieve a common goal⁴³

To date, it has not yet been possible to develop a strong artificial intelligence; there are still some unanswered questions. It is unclear whether a strong AI can gain its own consciousness and what role self-knowledge, empathy, memory and wisdom play in connection with the aforementioned properties. Another question is how to evaluate such a system from an ethical point of view.

At the point where a comprehensive and strong artificial intelligence is developed, it is assumed that there will be an artificial singularity.

According to Kurzweil, artificial singularity is understood as the point at which artificial systems rapidly improve themselves and thus accelerate technical progress to such an end that the future of humanity behind this event is no longer predictable. This is also said to be the last invention of mankind.⁴⁴

2.2.3 Algorithms and program

In order to produce an artificial intelligence, a computer algorithm is needed. An algorithm is used in mathematics in numerous cases, for example in the solution of a linear system of equations.⁴⁵

For this work, the following definition of an algorithm is used:

According to Dobler and Pomberger, an algorithm is a complete, precise and in a notation or language with exact definition, finite description of a step-by-step problem-solving method for determining searched data objects from given values of data objects. In an algorithm, each step consists of a number of executable, unique actions and an indication of the next step.⁴⁶

An algorithm is thus an instruction or a problem-solving procedure that shows how to solve a task. In order to be able to fulfill the instructions for action, certain characteristics are necessary. The algorithm must be universal, executable, efficient, understandable, unambiguous and correct. Furthermore, it must be finite.⁴⁷

Rimscha defines under generally valid that the algorithm fits not only to a specific problem, but to all similar tasks. For example, an algorithm is not about the question 'How do you sort several numbers?', but generally about 'How do you sort numbers?'. Furthermore, the algorithm must be executable. At last, many instructions must be given unambiguously, in a clear order, comprehensibly and effectively. There must be no room for interpretation and the algorithm must be efficient. For example, a statement must look like this: 'If the input value is straight, then go to the right, otherwise to the left.' The algorithm must come to a result, i.e. be finite.⁴⁸

The algorithm must also be correct and lead to the correct result. That is, he must always deliver a correct result for the particular problem.

An example of a simple algorithm is shown in Figure 8:

Abbildung in dieser Leseprobe nicht enthalten

Figure Example algorithm

By source: Scharwies/Kirst/Teufel (2019)

The scheme shown in Figure 8 is a simple algorithm. A decision diagram can be used to determine which of the two emeralds you own is a ruby and which is an emerald. The algorithm provides a solution to a specific problem. This process can be repeated as many times as desired.

The following table summarizes several authors; this is intended to represent some types of algorithms that have different characteristics and pursue different goals:

Type of algorithm

description

examples

recursive

A recursive algorithm means that in order to define a function, it itself is used again, but with different arguments. A recursive algorithm always requires a (non-recursive) initial condition or cancel condition.

Towers of Hanoi, Fibonacci

iterative

Algorithm that, in contrast to a recursive algorithm, proceeds gradually, i.e. iteratively. So only loops and branches are used, no self-appeals (recursions). Iteration is the repetition of structurally identical blocks by stringing them together.

Root calculation according to Heron of Alexandria

dynamic

Solving an optimization problem by dividing it into subproblems. Each partial solution of an optimal solution is itself an optimal solution to the respective sub-problem. (Dynamic programming)

Optimality principle of Bellman, CYK algorithm, Needleman desire algorithm

Heuristic

A heuristic algorithm is based on estimating, observing, guessing, or guessing. Heuristics are used to solve problems.

A* Search, Simulated Cooling, Evolutionary Algorithms and Fuzzy Logic

Determinis-tic

If the result of each statement is unique, and it is clear at every point in the flow which step to perform next, it is a deterministic algorithm.

Bubblesort, algorithm of Euclid

randomized (also stochastic or probabilistic)

Counterpart to the deterministic algorithm. Selection of random intermediate results in order to achieve a good or approximately correct result.

Las Vegas algorithms, Monte Carlo algorithms

Genetic/Evolutionary

An evolutionary algorithm is a stochastic search method that follows the principles of natural evolution. At the same time, a number of potential solutions are being worked on and selected according to the principle of 'the strongest survives'.

Modelling of various natural processes, such as recombination, crossover, selection, mutation, migration, competition, etc.

table : Explanation of different types of algorithms

By source: Mayer-Lindenberg (1998), p. 40 ff., Springer Fachmedien Wiesbaden GmbH (2013), p. 60 , Bellman (1957), p. 12 ff., Richter/Sander/Stucky (1993), p. 63 f., Hromkovič (2004), p. 66 ff. , Pohlheim (2000), p. 7 ff., Cormen et al. (2010), Mainzer (2019), pp. 3 f., 28, 61, 92, 232

According to Vöcking, algorithms are used to solve the following tasks:

- Search and sort Calculating, encrypting and encoding

- Planning, strategic action and computer simulation

- optimize⁴⁹

The path of a problem via the algorithm to its solution by a computer program is in the Figure 9: The path from the algorithm to the computer program depicted.

Abbildung in dieser Leseprobe nicht enthalten

Figure The path from algorithm to computer program

Source: Rimscha (2014), p. 4

As Figure 9 shows, the general problem is first described and a solution rule is developed for it. The resulting algorithm is translated into a problem-oriented language. The executability and the representation are precise, but machine-independent, predetermined. The problem-oriented program is automatically translated into a machine program by a translation algorithm (compiler). Then it is finally possible to solve concrete tasks and get a result.⁵⁰

Artificial intelligence consists of complex algorithms that are trained to make decisions independently.

2.2.4 Machine Learning and Big Data

Machine learning can be described as a field of research that deals with the computer-aided modeling and realization of learning phenomena.⁵¹

Figure 10 shows the comparison between traditional programming and machine learning.

Abbildung in dieser Leseprobe nicht enthalten

Figure Traditional programming and machine learning

Source:Gero Presser (2017)

In classical programming, the data and the algorithm are provided by the input of which a result is achieved. In machine learning, the data and the result are provided and the result is the algorithm. The algorithm is thus trained by input and the solution until it comes to the desired results and can automatically offer the solution for future inputs. It is crucial to go into more detail about the concept of learning.

Edelmann describes the term learning as a general, comprehensive term for changes in individual behavior to certain stimuli, signals, objects or situations. The changes are based on (repeated) experiences that are automatically registered and/or consciously processed. However, learning is only given if the change in behavior is not due to innate reaction tendencies (for example, reflexes), maturation processes or temporary changes in the state of an organism (for example, fatigue). The mediating processes of learning refer to changes in behavioral possibilities or willingness and form the latent basis for behavioral, comprehension and/or ways of thinking manifest in relation to the situation.⁵²

It can be narrowly defined that learning is an experience-based process that results in a permanent change in behavior or behavioral potential.⁵³

From this it can be concluded that machine learning is the ability of a machine or software to learn certain tasks that have been trained on the basis of experience (data). The knowledge no longer has to be coded and explicated by the software developer, but is acquired through learning.⁵⁴

According to Buxmann and others, machine learning can be divided into several types:

- Supervised Learning
- Unsupervised Learning
- Partially-and-semi-supervised learning
- Reinforcement Learning
- Active Learning

In the following, the different types of learning are explained in detail.

by the Supervised Learning the algorithm learns from given examples to recognize a relationship between the properties of the examples and the classes to which the examples belong. For example, the system is presented with animal images and the associated animal species. The system can then recognize patterns between certain characteristics (e.g. nose, legs, head) and class affiliation and apply them to new data. The decisive factor here is that the system determines the algorithm independently. It is a three-layer process. This involves a separation of the learning, evaluation and application phases.⁵⁵

by the Unsupervised Learning two aspects are distinguished. There is no separation between the different phases (learning, evaluation and application phases) and no classes are specified. The system thus creates classifications itself, according to which the input patterns are divided. For example, a system is shown a large number of animal images and these are grouped by the system into classes that have similar characteristics.

that Partially-and-semi-supervised learning is a mixture of supervised and unsupervised learning. Here, only a part of the inputs is classified.⁵⁶

by the Active Learning the aim is to improve the learning process of a system in an iterative process through targeted questions. The system thus has the option of requesting the correct output for some of the inputs.⁵⁷

by the Reinforcment-Learning learns the system through 'reward and punishment'. The system independently tests different strategies and adjusts the given variables so that the 'reward' is maximized. From the environment, the system receives direct feedback as to whether the action performed is positive or negative.⁵⁸

2.2.5 big data

According to Tenzer, big data acts as a driver and basic prerequisite for machine learning. The annual data growth rate is around 30%. In 2016, it was still about 16 zettabytes. By 2025, a data volume of about 163 zettabytes per year is forecast. For comparison: 1 zettabyte equals one billion terabytes. In the future, most of the data will not be generated by private users, but by companies.⁵⁹

[...]

---

¹ cf. Kiesewetter (2004), p. 15 ff.

² (Cf., ibid., p.77)

³ cf. Ibid., p. 56

⁴ cf. Reinheimer (2017), p. 49 ff.

⁵ cf. Hompel/Bauernhansl (2017), p. 155

⁶ cf. Schleer (2014)

⁷ cf. Zeit Online (2016)

⁸ cf. IBM -Deep Blue (2012); cf. IBM -Watson (2018)

⁹ cf. Mattke/online (2017)

¹⁰ cf. Council of Supply Chain Management Professionals (2013)

¹¹ cf. Kuhn/Hellingrath (2002), p. 10

¹² cf. Werner (2013), p. 5

¹³ (Cf., ibid., p.77)

¹⁴ cf. Porter (2008)

¹⁵ cf. Kuhn/Hellingrath (2002), p. 10 ff., cf. Werner (2013), p.6

¹⁶ cf. Kuhn/Hellingrath (2002), p. 10 ff.

¹⁷ Merten (2013)

¹⁸ cf. Klaus/Krieger (1998), p.14

¹⁹ cf. Werner (2013), p.8 f., cf. Klaus/Krieger/Krupp (2012), p. 4 f.

²⁰ Cf. the 6Rs of logistics Koether (2018) extended to 8Rs in the information age. cf. Hausladen (2016), p. 4 and modified by the author on 8R+

²¹ cf. Werner (2013), p. 25 f.

²² cf. Lenz (2008), p.31 f.

²³ cf. Kuhn/Hellingrath (2002), p. 16

²⁴ cf. Schulte (2017), p. 21 f.

²⁵ cf. Werner (2013), p. 29 ff.

²⁶ cf. Forrester (1958), p.37 ff.

²⁷ cf. Werner (2013), p. 40 f.

²⁸ cf. Simchi-Levi (2008), p.39 ff.

²⁹ cf. Werner (2013), p. 29 ff.

³⁰ (Cf., ibid., p.77)

³¹ cf. APICS (2017) , cf. Bolstorff/Rosenbaum/Poluha (2007), p. 15 f.

³² cf. Ziegenbein (2007), p. 10 f.

³³ (Cf., ibid., p.77)

³⁴ (Cf., ibid., p.77)

³⁵ Ibid., p. 56

³⁶ cf. McCarthy (1996)

³⁷ cf. Lackes/Siepermann (2018)

³⁸.cf. Kremp (2014)

³⁹ cf. Herbig (2018)

⁴⁰ cf. Lenzen (2002), p. 16

⁴¹ cf. Jaekel (2017), p. 21

⁴² cf. Buxmann/Schmidt (2018), p. 6

⁴³ cf. University of Oldenburg (undated)

⁴⁴ cf. Kurzweil (2015), p. 10 ff.

⁴⁵ cf. Menzel, W. (1991), p. 16

⁴⁶ cf. Pomberger/Dobler (2008), p.33

⁴⁷ (Cf., ibid., p.77)

⁴⁸ cf. Rimscha (2014), p. 3

⁴⁹ cf. Vöcking et al. (2008)

⁵⁰ cf. Menzel, W. (1991), p. 16

⁵¹ cf. Görz/Schmid/Schneeberger (2013), p. 406

⁵² cf. Edelmann (2012) cf. Fröhlich (2000), p. 282

⁵³ cf. Gerrig/Zimbardo/Graf (2008), p. 738

⁵⁴ cf. Buxmann/Schmidt (2018), p. 8 f.

⁵⁵ cf. Jannidis et al. (2017), p. 289 f.

⁵⁶ Cf. mindsquare GmbH (2019)

⁵⁷ cf. Reitmaier (2015), p. 15 f.

⁵⁸ cf. Gerlach (2011), p. 7

⁵⁹ cf. F. Tenzer (2019)

• Autor: Johannes Hangl (Autor:in)
• Erscheinungsjahr: 2020
• Seiten: 134
• Katalognummer: V1187321
• ISBN (PDF): 9783346603821
• Sprache: Englisch
• Schlagworte: artificial, intelligence, supply, chain, management, opportunities