This model visualizes gradient descent optimization - the fundamental algorithm used to train neural networks and other machine learning models. Agents represent different optimization algorithms searching for the minimum of a loss landscape (the “error surface” that ML models try to minimize during training).

The model demonstrates how different optimizer types (SGD, Momentum with different parameters) behave on various loss landscapes, from simple bowls to the notoriously difficult Rosenbrock “banana valley” function. This helps build intuition about why certain optimization algorithms work better than others for different problem geometries.

HOW IT WORKS

…

Although beneficial to scientific development, data sharing is still uncommon in many research areas. Various organisations, including funding agencies that endorse open science, aim to increase its uptake. However, estimating the large-scale implications of different policy interventions on data sharing by funding agencies, especially in the context of intense competition among academics, is difficult empirically. Here, we built an agent-based model to simulate the effect of different funding schemes (i.e., highly competitive large grants vs. distributive small grants), and varying intensity of incentives for data sharing on the uptake of data sharing by academic teams strategically adapting to the context.

The model’s aim is to represent the price dynamics under very simple market conditions, given the values adopted by the user for the model parameters. We suppose the market of a financial asset contains agents on the hypothesis they have zero-intelligence. In each period, a certain amount of agents are randomly selected to participate to the market. Each of these agents decides, in a equiprobable way, between proposing to make a transaction (talk = 1) or not (talk = 0). Again in an equiprobable way, each participating agent decides to speak on the supply (ask) or the demand side (bid) of the market, and proposes a volume of assets, where this number is drawn randomly from a uniform distribution. The granularity depends on various factors, including market conventions, the type of assets or goods being traded, and regulatory requirements. In some markets, high granularity is essential to capture small price movements accurately, while in others, coarser granularity is sufficient due to the nature of the assets or goods being traded

Inspired by the European project called GLODERS that thoroughly analyzed the dynamics of extortive systems, Bottom-up Adaptive Macroeconomics with Extortion (BAMERS) is a model to study the effect of extortion on macroeconomic aggregates through simulation. This methodology is adequate to cope with the scarce data associated to the hidden nature of extortion, which difficults analytical approaches. As a first approximation, a generic economy with healthy macroeconomics signals is modeled and validated, i.e., moderate inflation, as well as a reasonable unemployment rate are warranteed. Such economy is used to study the effect of extortion in such signals. It is worth mentioning that, as far as is known, there is no work that analyzes the effects of extortion on macroeconomic indicators from an agent-based perspective. Our results show that there is significant effects on some macroeconomics indicators, in particular, propensity to consume has a direct linear relationship with extortion, indicating that people become poorer, which impacts both the Gini Index and inflation. The GDP shows a marked contraction with the slightest presence of extortion in the economic system.

In a two-level hierarchical structure (consisting of the positions of managers and operators), persons holding these positions have a certain performance and the value of their own (personal perception in this, simplified, version of the model) perception of each other. The value of the perception of each other by agents is defined as a random variable that has a normal distribution (distribution parameters are set by the control elements of the interface).
In the world of the model, which is the space of perceptions, agents implement two strategies: rapprochement with agents that perceive positively and distance from agents that perceive negatively (both can be implemented, one of these strategies, or neither, the other strategy, which makes the agent stationary). Strategies are implemented in relation to those agents that are in the radius of perception (PerRadius).
The manager (Head) forms a team of agents. The performance of the group (the sum of the individual productivities of subordinates, weighted by the distance from the leader) varies depending on the position of the agents in space and the values of their individual productivities. Individual productivities, in the current version of the model, are set as a random variable distributed evenly on a numerical segment from 0 to 100. The manager forms the team 1) from agents that are in (organizational) radius (Op_Radius), 2) among agents that the manager perceives positively and / or negatively (both can be implemented, one of the specified rules, or neither, which means the refusal of the command formation).
Agents can (with a certain probability, given by the variable PrbltyOfDecisn%), in case of a negative perception of the manager, leave his group permanently.
It is possible in the model to change on the fly radii values, update the perception value across the entire population and the perception of an individual agent by its neighbors within the perception radius, and the probability values for a subordinate to make a decision about leaving the group.
You can also change the set of strategies for moving agents and strategies for recruiting a team manager. It is possible to add a randomness factor to the movement of agents (Stoch_Motion_Speed, the default is set to 0, that is, there are no random movements).
…

CPNorm is a model of a community of harvesters using a common pool resource where adhering to the optimal extraction level has become a social norm. The model can be used to explore the robustness of norm-driven cooperation in the commons.

Implemented as a virtual laboratory, this model explores transitions in land-use and livelihood decisions that emerge from changing local and global conditions.

The model’s purpose is to provide a potential explanation for the emergence, sustenance and decline of unpopular norms based on pluralistic ignorance on a social network.

Schelling and Sakoda prominently proposed computational models suggesting that strong ethnic residential segregation can be the unintended outcome of a self-reinforcing dynamic driven by choices of individuals with rather tolerant ethnic preferences. There are only few attempts to apply this view to school choice, another important arena in which ethnic segregation occurs. In the current paper, we explore with an agent-based theoretical model similar to those proposed for residential segregation, how ethnic tolerance among parents can affect the level of school segregation. More specifically, we ask whether and under which conditions school segregation could be reduced if more parents hold tolerant ethnic preferences. We move beyond earlier models of school segregation in three ways. First, we model individual school choices using a random utility discrete choice approach. Second, we vary the pattern of ethnic segregation in the residential context of school choices systematically, comparing residential maps in which segregation is unrelated to parents’ level of tolerance to residential maps reflecting their ethnic preferences. Third, we introduce heterogeneity in tolerance levels among parents belonging to the same group. Our simulation experiments suggest that ethnic school segregation can be a very robust phenomenon, occurring even when about half of the population prefers mixed to segregated schools. However, we also identify a “sweet spot” in the parameter space in which a larger proportion of tolerant parents makes the biggest difference. This is the case when parents have moderate preferences for nearby schools and there is only little residential segregation. Further experiments are presented that unravel the underlying mechanisms.

The FRAMe (Flood Resilience Agent-Based Model) serves as a framework designed to simulate flood resilience dynamics at the community level, focusing on a rural settlement in the Mekong River Basin. Integrating empirical data from extensive surveys, Bayesian networks, and hydrological simulations, the framework quantifies resilience as a trade-off between robustness (resistance to damage) and adaptability (capacity for dynamic response). Agents include households, governments, and other actors, linked by social and governance networks that facilitate knowledge transfer, resource distribution, and risk communication. FRAMe incorporates mechanisms for flood forecasting, policy interventions (education, aid, insurance), and individual and collective decision-making, grounded in Protection Motivation Theory and MoHuB frameworks. The framework’s spatially explicit design leverages GIS data, which supports scenario testing of governance structures and stakeholder interactions. By examining policy scenarios and agent behavior, FRAMe aims to inform adaptive flood management strategies and enhance community resilience.