Mark R. Kramer

A framework for early warning and proactive control systems in food supply chain networks

It is inherent to food supply chain networks that performance deviations occur occasionally due to variations in product quality and quantity. To reduce losses, one wants to be informed about such deviations as soon as possible, preferably even before they occur. Then it is possible to take actions to prevent or reduce negative consequences. In practice, extensive operational data is recorded, forming a valuable source for early warning and proactive control systems, i.e. decision support systems for prediction and prevention of such performance problems. Data mining methods are ideal for analyzing such data sources and extracting useable information from them. In this paper, we present a novel framework for early warning and proactive control systems that combine expert knowledge and data mining methods to exploit recorded data. We discuss the implementation of a prototype system and the experiences from a case study regarding the applicability of the framework.

A chain game for distributed trading and negotiation

This position paper introduces a simulation gaming environment for enacting a production network. The environment aims to be an integrative laboratory for investigating supply networks, as well as being a versatile training tool. The primary focus is on food production networks. The environment enables a number of teams of participants, each representing one actor in a food chain, to conduct business together. The teams can have the role of auction, co-operation, wholesaler, factory, retail chain, and retail outlet. Producers and consumers are either enacted or simulated. The game leaders freely determine the products and production methods in each run of the game. The gaming environment takes performance, process and institutional aspects of chains into account. It is particularly suited for investigating issues of sustainability and trust. Currently the gaming environment is under development. The paper presents version 1B. This version can be found at http://www.chaingame.org. It runs on the Web, enabling to model distributed chains.

An Agent-Based Information Management Model of the Chinese Pig Sector

This paper investigates the effect of a selected top-down measure (what-if scenario) on actual agent behaviour and total system behaviour by means of an agent-based simulation model, when agents’ behaviour cannot fully be managed because the agents are autonomous. The Chinese pork sector serves as case. A multi-level perspective is adopted: the top-down information management measures for improving pork quality, the variation in individual farmer behaviour, and the interaction structures with supply chain partners, governmental representatives and peer farmers. To improve quality, farmers need information, which they can obtain from peers, suppliers and government. Satisfaction or dissatisfaction with their personal situation initiates change of behaviour. Aspects of personality and culture affect the agents’ evaluations, decisions and actions. Results indicate that both incentive (demand) and the possibility to move (quality level within reach) on farmer level are requirements for an increase of total system quality. A more informative governmental representative enhances this effect.

Multi-dimensional information diffusion and balancing market supply: an agent-based approach

This agent-based information management model is designed to explore how multi-dimensional information, spreading through a population of agents (for example farmers) affects market supply. Farmers make quality decisions that must be aligned with available markets. Markets distinguish themselves by means of requirements which are expressed over multiple quality dimensions. In order to supply at a market, a supplier’s information should match the market’s requirements. Information diffusion is affected by network structure among agents, and by information turnover. Research questions concern the effect of information turnover and network structure on market supply. Results show that there is a huge effect of information turnover. The percentage of suppliers having to resort to the dump market decreases when information supply rate ISR and average number of friends NFR increase. The higher the values of ISRR and NFR, the higher the percentage of suppliers able to reach high markets. There is an influence of network structure: the more connections, the better the results with respect to market supply, but the nature of these connection seems to be of lesser importance. Contrary to our expectations, there is hardly an effect of network topology. With sufficient information in the system, differences in diffusion process appear to be not significant.

An Agent-Based Information Management Approach to Smoothen the Pork Cycle in China

The objective of our research is to study the relationship between (a) the spread of information at farmer level and (b) the emerging behaviour at sector level, applied to the case of the pork cycle in China, using an agent-based model. For this paper, we investigate the effect of farmers’ individual supply decisions on the overall supply pattern in the sector. We apply a basic agent-based supply- and demand model populated with pig farmers, where supply is based on price expectations that include a time lag. The farmers decide upon their future supply (at farm level) using the price expectations they are able to make based on the information at their disposal.We compare our agent-based model with the classical cobweb model, which exhibits periodical over- and under-supply. This periodicity is not desirable, as is illustrated by a realistic example from the pork sector in China. The Chinese government tries to smoothen the overall supply and demand pattern by acting as a speculator as soon as price imbalance at total system level exceeds a threshold value, hence intervening at system level. Our agent-based model displays similar periodicity is also possible at individual level. An emergent result from the comparison is that mapping of economic supply and demand functions to individual agents’ decisions is not straightforward. Our model is a fruitful basis for further research, which will include social interaction, imitation behaviour and a more sophisticated information diffusion process that reflects the rate at which a farmers population adopts information.

Sustainable animal welfare: does forcing farmers into transition help?

Dutch society is changing, and so is its attitude towards animal welfare. Meat retailers respond by laying down minimum-quality criteria for farmers who wish to supply to supermarkets—forcing them to either aim for higher quality or lose their market. Policy-wise this is a top-down measure that leads to a redistribution of markets. From farmer perspective, a transition with more individual freedom to adapt seems more sustainable. By means of an existing agent-based model, this paper investigates two policies for such a market switch: immediate transition—‘sudden death’ (SD)—versus gradual change—‘graceful degradation’ (GD). Both farmers and available markets are modelled as agents. Each farmer has a collection of multi-dimensional information items, under certain conditions exchangeable with other farmers in his network, representing his knowledge and skills. Information items are a farmer’s key to the market, as market criteria are expressed in terms of information requirements. We tested the effect of SD and GD policies on market redistribution, varying markets sets, available information, and network size. Results show that policy does not matter for final market redistribution, but that GD policy indeed allows more farmers to keep away from poverty, especially in information-poor situations. With GD, we see a temporarily higher inequality of income distribution over individuals (Gini) worth exploring. Studying transitions with respect to both individuals and the system as a whole may be promising for other domains as well. The model is applicable to any situation that implies aligning heterogeneous suppliers with a multi-dimensional demand spectrum.

Influence of Losing Multi-dimensional Information in an Agent-Based Model

This agent-based study investigates the effect of losing information on market performance of agents in a marketplace with various quality requirements. It refines an existing model on multi-dimensional information diffusion among agents in a network. The agents need to align their supply with available markets, the quality criteria of which must match the agents’ information. Turnover (information entering and leaving the system) had a significant effect in the old model. Information items became obsolete based on age, causing a risk for the agents to lose valuable information. In the refined model presented here, an information item may become obsolete based on two additional aspects: (1) whether it is ‘in use’ for meeting the agent’s current market criteria, and (2) its value, reflecting its owner’s experience or skill with the information item. The research questions concern the influence of these two aspects on model outcomes. Two key parameters are value-threshold, below which items are candidate for disposal, and keep-chance, indicating the probability that in-use items are not disposed of. Both simulation runs and a local sensitivity analysis were performed. Simulation results show that value-threshold is a more influential parameter than keep-chance. An interesting pattern suggesting a tipping point was observed: with increasing value-threshold, agents initially reach higher quality, but then the quality diminishes again. This pattern is consistently observed for the majority of parameter settings. An explanation is that agents with only high-valued information cannot afford to lose anything. The sensitivity analysis adds insight to where keep-chance and value-threshold are most influential, and where other parameters are responsible for observed outputs. The sensitivity analysis does not provide any further insight in why the observed tipping point occurs. The paper also aims to highlight methodological issues with respect to refining an existing model in such a way that results of successive model versions are still comparable, and observed differences can be attributed to newly introduced changes.

The status–power arena: a comprehensive agent-based model of social status dynamics and gender in groups of children

Despite the urgency of this issue, AI still struggles to represent social life. This article presents a comprehensive agent-based model that investigates status-power dynamics in groups. Kemper’s sociological status–power theory of social relationships, and a literature review on school children in middle youth, is its basis. The model allows us to investigate causation of the near-ubiquitous phenomenon that females have lower social status on average than males. Possible causes included in the model are children’s dispositional traits (kindness, beauty, and physical power), schoolyard culture (social acceptability of fighting), behavioural strategy (amount of rough-and-tumble play) and the balance between public and dyadic sources of status. An agent-based model of a virtual schoolyard was created in which the children assemble in changing groups and mutually confer status. The status conferred upon a child modifies the status it holds. Rough-and-tumble is modelled as ambiguous: it is intended as a status conferral, but may be perceived as a power move. Running many trials of the model we found that in time, depending on the parameter settings, a gender-based status gap emerged. Rough-and-tumble play had more impact on emergent status differences than did physical power differences. Social acceptability of fighting also strongly moderated the resulting status gap. Placing more weight on dyadic relationship could alleviate status loss. All these model behaviours are in line with empirical findings of child behaviour studies at schools. They have face validity for social status issues in the adult world. We conclude from this that this kind of agent-based model merits use in studying the status–power dynamics of other issues in child behaviour, or indeed in social behaviour in general.

Managing Variant Calling Files the Big Data Way: Using HDFS and Apache Parquet

Big Data has been seen as a remedy for the efficient management of the ever-increasing genomic data. In this paper, we investigate the use of Apache Spark to store and process Variant Calling Files (VCF) on a Hadoop cluster. We demonstrate Tomatula, a software tool for converting VCF files to Apache Parquet storage format, and an application to query variant calling datasets. We evaluate how the wall time (i.e. time until the query answer is returned to the user) scales out on a Hadoop cluster storing VCF files, either in the original flat-file format, or using the Apache Parquet columnar storage format. Apache Parquet can compress the VCF data by around a factor of 10, and supports easier querying of VCF files as it exposes the field structure. We discuss advantages and disadvantages in terms of storage capacity and querying performance with both flat VCF files and Apache Parquet using an open plant breeding dataset. We conclude that Apache Parquet offers benefits for reducing storage size and wall time, and scales out with larger datasets.