NY
N. Yorke-Smith
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
122 records found
1
...
Efficiently assigning vessels to berths in a busy port is a hard scheduling problem: decisions must be made in real time, under uncertainty about future arrivals, and with the goal of minimizing how long ships spend in the port. We present a learning-based approach in which an agent is trained, purely through trial and error, to perform these assignments one at a time as the vessels arrive. The agent represents the current port state as a graph, capturing relationships between waiting vessels, available berths, and ships still on their way, and learns which assignments are promising via graph-based attention. Our formulation incorporates vessel priorities, berth-dependent handling times, and a skip action that allows the agent to postpone assignments when beneficial. Across a range of port congestion levels, the trained agent outperforms classical heuristic rules, achieving on average a 20% reduction in priority-weighted turnaround cost relative to the best heuristic baseline.
...
Efficiently assigning vessels to berths in a busy port is a hard scheduling problem: decisions must be made in real time, under uncertainty about future arrivals, and with the goal of minimizing how long ships spend in the port. We present a learning-based approach in which an agent is trained, purely through trial and error, to perform these assignments one at a time as the vessels arrive. The agent represents the current port state as a graph, capturing relationships between waiting vessels, available berths, and ships still on their way, and learns which assignments are promising via graph-based attention. Our formulation incorporates vessel priorities, berth-dependent handling times, and a skip action that allows the agent to postpone assignments when beneficial. Across a range of port congestion levels, the trained agent outperforms classical heuristic rules, achieving on average a 20% reduction in priority-weighted turnaround cost relative to the best heuristic baseline.
Book chapter
(2026)
-
Pengwei Guo, Noortje Wagemakers, Sandra Barbosa Nunes, Neil Yorke-Smith, Virginie Wiktor
Mixing torque reflects the interaction between the mixer and fresh mortar, providing insights into material consistency. Traditionally, obtaining torque measurements requires specialised sensors integrated into mixers, which adds cost and limits their practicality for large-scale or on-site use. To address this, this study proposes a deep learning framework that predicts real-time torque values directly from mixing videos. Instead of relying on specialised sensors or equipment, the model extracts spatial and temporal features from consecutive video frames using a time-series architecture. Specifically, a hybrid ResNet–LSTM model is employed: ResNet encodes spatial features from each individual frame, while the LSTM captures temporal dependencies across sequences of frames. This allows the model to learn how visual changes in the mixing process correlate with the evolving torque. A dataset comprising 21 mortar mixtures with varying compositions was collected, including synchronised video footage and torque measurements recorded throughout the mixing period. Workability, flexural and compressive strength tests were performed after mixing. The model achieved R2 scores of 0.992 (training), 0.989 (validation), and 0.936 (testing), indicating that the model achieved high accuracy with strong generalisation ability across unseen data. The inference time is under 60 ms per 5-frame sequence. The proposed method enables fast, non-contact, and reliable torque estimation, offering a practical solution for intelligent monitoring of mixing processes in real-world settings.
...
Mixing torque reflects the interaction between the mixer and fresh mortar, providing insights into material consistency. Traditionally, obtaining torque measurements requires specialised sensors integrated into mixers, which adds cost and limits their practicality for large-scale or on-site use. To address this, this study proposes a deep learning framework that predicts real-time torque values directly from mixing videos. Instead of relying on specialised sensors or equipment, the model extracts spatial and temporal features from consecutive video frames using a time-series architecture. Specifically, a hybrid ResNet–LSTM model is employed: ResNet encodes spatial features from each individual frame, while the LSTM captures temporal dependencies across sequences of frames. This allows the model to learn how visual changes in the mixing process correlate with the evolving torque. A dataset comprising 21 mortar mixtures with varying compositions was collected, including synchronised video footage and torque measurements recorded throughout the mixing period. Workability, flexural and compressive strength tests were performed after mixing. The model achieved R2 scores of 0.992 (training), 0.989 (validation), and 0.936 (testing), indicating that the model achieved high accuracy with strong generalisation ability across unseen data. The inference time is under 60 ms per 5-frame sequence. The proposed method enables fast, non-contact, and reliable torque estimation, offering a practical solution for intelligent monitoring of mixing processes in real-world settings.
On data-driven robust optimization with multiple uncertainty subsets
Unified uncertainty set representation and mitigating conservatism
Constructing uncertainty sets as unions of multiple subsets has emerged as an effective approach for creating compact and flexible uncertainty representations in data-driven robust optimization (RO). This paper focuses on two separate research questions. The first concerns the computational challenge in applying these uncertainty sets in RO-based predictive control. To address this, a monolithic mixed-integer representation of the uncertainty set is proposed to uniformly describe the union of multiple subsets, enabling the computation of the worst-case uncertainty scenario across all subsets within a single mixed-integer linear programming (MILP) problem. The second research question focuses on mitigating the conservatism of conventional RO formulations by leveraging the structure of the uncertainty set. To achieve this, a novel objective function is proposed to exploit the uncertainty set structure and integrate the existing RO and distributionally robust optimization (DRO) formulations, yielding less conservative solutions than conventional RO formulations, while avoiding the high-dimensional continuous uncertainty distributions and the high computational burden typically associated with existing DRO formulations. Given the proposed formulations, numerically efficient computation methods based on column-and-constraint generation (CCG) are also developed. Extensive simulations across three case studies are performed to demonstrate the effectiveness of the proposed schemes.
...
Constructing uncertainty sets as unions of multiple subsets has emerged as an effective approach for creating compact and flexible uncertainty representations in data-driven robust optimization (RO). This paper focuses on two separate research questions. The first concerns the computational challenge in applying these uncertainty sets in RO-based predictive control. To address this, a monolithic mixed-integer representation of the uncertainty set is proposed to uniformly describe the union of multiple subsets, enabling the computation of the worst-case uncertainty scenario across all subsets within a single mixed-integer linear programming (MILP) problem. The second research question focuses on mitigating the conservatism of conventional RO formulations by leveraging the structure of the uncertainty set. To achieve this, a novel objective function is proposed to exploit the uncertainty set structure and integrate the existing RO and distributionally robust optimization (DRO) formulations, yielding less conservative solutions than conventional RO formulations, while avoiding the high-dimensional continuous uncertainty distributions and the high computational burden typically associated with existing DRO formulations. Given the proposed formulations, numerically efficient computation methods based on column-and-constraint generation (CCG) are also developed. Extensive simulations across three case studies are performed to demonstrate the effectiveness of the proposed schemes.
This article proposes an evidence-based policy recommendation framework integrating social media data and natural language processing methods, to support inclusive and efficient transport policy-making. Given that current research underscores the crucial role of both external and psychological variables in individual travel decisions, psychological features – such as beliefs, attitudes or values – are frequently used as latent variables for travel behaviour interpretation and travel choice modelling. However, user-centric policy recommendations based on dynamic psychological variables are still limited. Most studies rely on survey data, which neglects the urgent dynamic trend of user perception change and its underlying relationship with travel behaviour. Hence there is a lack of illustration on how these psychological variables can be further used at specific temporal and spatial levels for travel behaviour interpretation. This would be valuable to identify priorities for more targeted (sustainability and other) policies and interventions. In this article, we utilize sentiment analysis and dynamic topic modelling to represent the spatial–temporal variance of psychological features. Integrating with corresponding travel behaviour, we illustrate how these dynamic psychological features can distinguish travel dissonance, identify key motivations, and reflect urgent social demands at precise spatial–temporal levels. We demonstrate these advances in a case study in New York City from 2019 to 2022 using Twitter (X) data. A comparison with existing travel-related policies in the case study validates the feasibility of our framework to support evidence-based policy recommendations. We conclude by discussing the potential of this framework to support sustainable transport promotion.
...
This article proposes an evidence-based policy recommendation framework integrating social media data and natural language processing methods, to support inclusive and efficient transport policy-making. Given that current research underscores the crucial role of both external and psychological variables in individual travel decisions, psychological features – such as beliefs, attitudes or values – are frequently used as latent variables for travel behaviour interpretation and travel choice modelling. However, user-centric policy recommendations based on dynamic psychological variables are still limited. Most studies rely on survey data, which neglects the urgent dynamic trend of user perception change and its underlying relationship with travel behaviour. Hence there is a lack of illustration on how these psychological variables can be further used at specific temporal and spatial levels for travel behaviour interpretation. This would be valuable to identify priorities for more targeted (sustainability and other) policies and interventions. In this article, we utilize sentiment analysis and dynamic topic modelling to represent the spatial–temporal variance of psychological features. Integrating with corresponding travel behaviour, we illustrate how these dynamic psychological features can distinguish travel dissonance, identify key motivations, and reflect urgent social demands at precise spatial–temporal levels. We demonstrate these advances in a case study in New York City from 2019 to 2022 using Twitter (X) data. A comparison with existing travel-related policies in the case study validates the feasibility of our framework to support evidence-based policy recommendations. We conclude by discussing the potential of this framework to support sustainable transport promotion.
Corruption is a familiar and pressing problem in the performance of administrative bureaucracies. Changing the organisational structure is one way ventured to combat corrupt practices within a hierarchical organisation. Previous works have studied organisational change from various lenses, including equation-based modelling. We address the question of what level of hierarchy is optimal in such an organisation by means of agent-based simulation. We argue that agent-based models are uniquely suited for the exploratory modelling of corruption due to their capturing of localised, individualised behaviours. Our preliminary findings are that a less hierarchical organisational structure: 1) tend to lead to less corrupt acts committed, and 2) tends to lead to more societal welfare generated – however, 3) less corruption and more societal welfare do not always go hand in hand. We begin to reconcile these seemingly paradoxical results using theories from developmental economics.
...
Corruption is a familiar and pressing problem in the performance of administrative bureaucracies. Changing the organisational structure is one way ventured to combat corrupt practices within a hierarchical organisation. Previous works have studied organisational change from various lenses, including equation-based modelling. We address the question of what level of hierarchy is optimal in such an organisation by means of agent-based simulation. We argue that agent-based models are uniquely suited for the exploratory modelling of corruption due to their capturing of localised, individualised behaviours. Our preliminary findings are that a less hierarchical organisational structure: 1) tend to lead to less corrupt acts committed, and 2) tends to lead to more societal welfare generated – however, 3) less corruption and more societal welfare do not always go hand in hand. We begin to reconcile these seemingly paradoxical results using theories from developmental economics.
This extended abstract considers the gap between the theoretical ideals and practical realities of distributed and money-free resource exchange protocols among self-interested agents. We sketch a convex program that corresponds to the centralised exchange market equilibrium. Secondly we state the convergence of proportional exchange allocation, comprising a distributed solution to the equilibrium. Thirdly, summarising a simulation study, we observe that distributed, mixed-strategy protocols can achieve stable and desirable outcomes, modulated by sensitivity to population diversity, limited information, and strategic agent behaviour.
...
This extended abstract considers the gap between the theoretical ideals and practical realities of distributed and money-free resource exchange protocols among self-interested agents. We sketch a convex program that corresponds to the centralised exchange market equilibrium. Secondly we state the convergence of proportional exchange allocation, comprising a distributed solution to the equilibrium. Thirdly, summarising a simulation study, we observe that distributed, mixed-strategy protocols can achieve stable and desirable outcomes, modulated by sensitivity to population diversity, limited information, and strategic agent behaviour.
Conference paper
(2026)
-
Robin Manhaeve, Francesco Giannini, Mehdi Ali, Damiano Azzolini, Alice Bizzarri, Andrea Borghesi, Samuele Bortolotti, Sebastijan Dumančić, Neil Yorke-Smith, More authors...
Neural-symbolic (NeSy) AI has gained a lot of popularity by enhancing learning models with explicit reasoning capabilities. Both new systems and new benchmarks are constantly introduced and used to evaluate learning and reasoning skills. The large variety of systems and benchmarks, however, makes it difficult to establish a fair comparison among the various frameworks, let alone a unifying set of benchmarking criteria. This paper analyzes the state-of-the-art in benchmarking NeSy systems, studies its limitations, and proposes ways to overcome them. We categorize popular neural-symbolic frameworks into three groups: model-theoretic, proof-theoretic fuzzy, and proof-theoretic probabilistic systems. We show how these three categories have distinct strengths and weaknesses, and how this is reflected in the type of tasks and benchmarks to which they are applied.
...
Neural-symbolic (NeSy) AI has gained a lot of popularity by enhancing learning models with explicit reasoning capabilities. Both new systems and new benchmarks are constantly introduced and used to evaluate learning and reasoning skills. The large variety of systems and benchmarks, however, makes it difficult to establish a fair comparison among the various frameworks, let alone a unifying set of benchmarking criteria. This paper analyzes the state-of-the-art in benchmarking NeSy systems, studies its limitations, and proposes ways to overcome them. We categorize popular neural-symbolic frameworks into three groups: model-theoretic, proof-theoretic fuzzy, and proof-theoretic probabilistic systems. We show how these three categories have distinct strengths and weaknesses, and how this is reflected in the type of tasks and benchmarks to which they are applied.
How ex ante policy evaluation supports circular city development
Amsterdam's mass timber construction policy
Journal article
(2025)
-
Felipe Bucci Ancapi, Marvin Kleijweg, Karel Van den Berghe, Neil Yorke-Smith, Ellen van Bueren
This article aimed to assess the potential impact of policy actions to support mass timber construction through an ex ante policy analysis in Amsterdam. Through a combination of policy coherence analysis and agent-based simulation, the study evaluates 130 policy actions, including 80 specific instruments, for the transition from traditional masonry to mass timber construction. The coherence analysis reveals a predominance of regulatory instruments (62%) and a lack of active economic measures (16%), which limits their impact on circular city development. The simulation tested three instruments - demolition notification, a mass timber subsidy proxy and a carbon tax proxy - to assess their individual and combined effectiveness. Isolated measures, such as material price adjustments, were found to be insufficient due to systemic inertia. However, the combination of subsidies and carbon taxes proves more effective, significantly increasing the uptake of mass timber construction as its cost is reduced and construction companies develop expertise. A key finding highlights the complementary role of recycled concrete in supporting mass timber construction, highlighting the need for integrated policies targeting both mass timber and secondary materials. Improving industry knowledge and expertise is identified as a transformative approach to reducing costs and overcoming barriers to adoption. This research is the first contribution to demonstrate the value of ex ante policy evaluation and agent-based simulation in formulating coherent and effective policies for circular city transitions. Policy makers in Amsterdam and other Dutch cities are advised to implement synergistic instruments, support local material reuse and invest in capacity building to achieve carbon neutrality and resource circularity in urban construction. The findings provide actionable guidance for Amsterdam and similar cities seeking to promote sustainable and circular urban environments.
...
This article aimed to assess the potential impact of policy actions to support mass timber construction through an ex ante policy analysis in Amsterdam. Through a combination of policy coherence analysis and agent-based simulation, the study evaluates 130 policy actions, including 80 specific instruments, for the transition from traditional masonry to mass timber construction. The coherence analysis reveals a predominance of regulatory instruments (62%) and a lack of active economic measures (16%), which limits their impact on circular city development. The simulation tested three instruments - demolition notification, a mass timber subsidy proxy and a carbon tax proxy - to assess their individual and combined effectiveness. Isolated measures, such as material price adjustments, were found to be insufficient due to systemic inertia. However, the combination of subsidies and carbon taxes proves more effective, significantly increasing the uptake of mass timber construction as its cost is reduced and construction companies develop expertise. A key finding highlights the complementary role of recycled concrete in supporting mass timber construction, highlighting the need for integrated policies targeting both mass timber and secondary materials. Improving industry knowledge and expertise is identified as a transformative approach to reducing costs and overcoming barriers to adoption. This research is the first contribution to demonstrate the value of ex ante policy evaluation and agent-based simulation in formulating coherent and effective policies for circular city transitions. Policy makers in Amsterdam and other Dutch cities are advised to implement synergistic instruments, support local material reuse and invest in capacity building to achieve carbon neutrality and resource circularity in urban construction. The findings provide actionable guidance for Amsterdam and similar cities seeking to promote sustainable and circular urban environments.
Noise pollution from heat pumps (HPs) has been an emerging concern to their broader adoption, especially in densely populated areas. This paper explores a model predictive control (MPC) approach for climate control of buildings, aimed at minimizing the noise nuisance generated by HPs. By exploiting a piecewise linear approximation of HP noise patterns and assuming linear building thermal dynamics, the proposed design can be generalized to handle various HP acoustic patterns with mixed-integer linear programming (MILP). Additionally, two computationally efficient options for defining the noise cost function in the proposed MPC design are discussed. Numerical experiments on a high-fidelity building simulator are performed to demonstrate the viability and effectiveness of the proposed design. Simulation results show that minimizing the excess of HP noise over ambient noise is effective in mitigating the HP noise nuisance. Further, compared with the conventional MPC-based building climate control scheme, the proposed approach can effectively reduce the HP noise pollution with only a minor energy cost increase.
...
Noise pollution from heat pumps (HPs) has been an emerging concern to their broader adoption, especially in densely populated areas. This paper explores a model predictive control (MPC) approach for climate control of buildings, aimed at minimizing the noise nuisance generated by HPs. By exploiting a piecewise linear approximation of HP noise patterns and assuming linear building thermal dynamics, the proposed design can be generalized to handle various HP acoustic patterns with mixed-integer linear programming (MILP). Additionally, two computationally efficient options for defining the noise cost function in the proposed MPC design are discussed. Numerical experiments on a high-fidelity building simulator are performed to demonstrate the viability and effectiveness of the proposed design. Simulation results show that minimizing the excess of HP noise over ambient noise is effective in mitigating the HP noise nuisance. Further, compared with the conventional MPC-based building climate control scheme, the proposed approach can effectively reduce the HP noise pollution with only a minor energy cost increase.
Uncertainty quantification remains a difficult challenge in reinforcement learning. Several algorithms exist that successfully quantify uncertainty in a practical setting. However it is unclear whether these algorithms are theoretically sound and can be expected to converge. Furthermore, they seem to treat the uncertainty in the target parameters in different ways. In this work, we unify several practical algorithms into one theoretical framework by defining a new Bellman operator on distributions, and show that this Bellman operator is a contraction. We highlight use cases of our framework by analyzing an existing Bayesian Q-learning algorithm, and also introduce a novel uncertainty-aware variant of PPO that adaptively sets its clipping hyperparameter.
...
Uncertainty quantification remains a difficult challenge in reinforcement learning. Several algorithms exist that successfully quantify uncertainty in a practical setting. However it is unclear whether these algorithms are theoretically sound and can be expected to converge. Furthermore, they seem to treat the uncertainty in the target parameters in different ways. In this work, we unify several practical algorithms into one theoretical framework by defining a new Bellman operator on distributions, and show that this Bellman operator is a contraction. We highlight use cases of our framework by analyzing an existing Bayesian Q-learning algorithm, and also introduce a novel uncertainty-aware variant of PPO that adaptively sets its clipping hyperparameter.
Crew costs make up the second largest expense for airlines, behind only fuel costs. This motivates a potential gain in improving crew efficiency within the bounds set by the law and collective labour agreements. Doing so requires to take into account aircraft routes and crew pairings, and the specifics of the airline’s network. This work presents an integrated model for obtaining efficient crew pairings for airlines operating point-to-point networks, while also allowing for flight retiming. By considering simultaneously both crew pairing and constrained aircraft routing, better-performing solutions can be obtained. The greater complexity of the integrated model is addressed by means of a custom branch-and-price approach with a shortest path pricing sub-problem, in order to obtain exact solutions. The results of the integrated model are evaluated on a real-world case of an European low-cost carrier that operates a short-haul point-to-point network. Results show a reduction in crew duties of 10% and an increase in crew efficiency metrics by up to 1.5%, optimising the carrier’s complete network of 926 flights over a full week.
...
Crew costs make up the second largest expense for airlines, behind only fuel costs. This motivates a potential gain in improving crew efficiency within the bounds set by the law and collective labour agreements. Doing so requires to take into account aircraft routes and crew pairings, and the specifics of the airline’s network. This work presents an integrated model for obtaining efficient crew pairings for airlines operating point-to-point networks, while also allowing for flight retiming. By considering simultaneously both crew pairing and constrained aircraft routing, better-performing solutions can be obtained. The greater complexity of the integrated model is addressed by means of a custom branch-and-price approach with a shortest path pricing sub-problem, in order to obtain exact solutions. The results of the integrated model are evaluated on a real-world case of an European low-cost carrier that operates a short-haul point-to-point network. Results show a reduction in crew duties of 10% and an increase in crew efficiency metrics by up to 1.5%, optimising the carrier’s complete network of 926 flights over a full week.
As a tool serving other disciplines of enquiry, artificial intelligence (AI) offers the potential of a potent discovery, a design and analysis paradigm to address (new) questions in urban planning. This thematic issue raises a forum for cross-disciplinary dialogues at the intersection of urban planning and AI. Nine articles discuss both emerging use cases in urban planning practice and the relevant AI techniques being used and developed, as well as articulate the challenges associated. Future development of AI in urban planning shall address the ethical, inclusive, and just implications of AI applications for urban planning while navigating human and AI agents’ interactions and intra-actions to facilitate a better understanding of the intentions of AI development and use, and the impacts on the behaviour of designers and users in complex urban planning practices.
...
As a tool serving other disciplines of enquiry, artificial intelligence (AI) offers the potential of a potent discovery, a design and analysis paradigm to address (new) questions in urban planning. This thematic issue raises a forum for cross-disciplinary dialogues at the intersection of urban planning and AI. Nine articles discuss both emerging use cases in urban planning practice and the relevant AI techniques being used and developed, as well as articulate the challenges associated. Future development of AI in urban planning shall address the ethical, inclusive, and just implications of AI applications for urban planning while navigating human and AI agents’ interactions and intra-actions to facilitate a better understanding of the intentions of AI development and use, and the impacts on the behaviour of designers and users in complex urban planning practices.
Future energy markets for low voltage AC and DC distribution systems will facilitate prosumer participation in the market. To comply with market regulations and grid constraints, a tailored market design reflecting (DC) operational requirements is needed. Our previous work identified a locational energy market design. However, its real-life implementation faces challenges due to uncertainties in system operation, prosumer preferences, and bidding strategies. This article tests the market design under uncertain scenarios. To this end, we develop an agent-based model that simulates typical electric vehicle user preferences and bidding strategies, influenced by varying degrees of range anxiety. The market design is tested in challenging scenarios with a high share of solar panels and electric vehicles, modelled using the high-resolution Pecan Street database. Simulations indicate that the proposed market design maintains both economic efficiency and system reliability under real-life uncertainties. This in turn indicates the practical feasibility of locational energy markets in helping to integrate renewable generation sources and bidirectional power flows.
...
Future energy markets for low voltage AC and DC distribution systems will facilitate prosumer participation in the market. To comply with market regulations and grid constraints, a tailored market design reflecting (DC) operational requirements is needed. Our previous work identified a locational energy market design. However, its real-life implementation faces challenges due to uncertainties in system operation, prosumer preferences, and bidding strategies. This article tests the market design under uncertain scenarios. To this end, we develop an agent-based model that simulates typical electric vehicle user preferences and bidding strategies, influenced by varying degrees of range anxiety. The market design is tested in challenging scenarios with a high share of solar panels and electric vehicles, modelled using the high-resolution Pecan Street database. Simulations indicate that the proposed market design maintains both economic efficiency and system reliability under real-life uncertainties. This in turn indicates the practical feasibility of locational energy markets in helping to integrate renewable generation sources and bidirectional power flows.
Solvers for constraint optimisation problems exploit variable and value ordering heuristics. Numerous expert-designed heuristics exist, while recent research learns novel, customised heuristics from past problem instances. This article addresses unseen problems for which no historical data is available. We propose one-shot learning of customised, problem instance-specific heuristics. To do so, we introduce the concept of deep heuristics, a data-driven approach to learn extended versions of a given variable ordering heuristic online. First, for a problem instance, an initial online probing phase collects data, from which a deep heuristic function is learned. The learned heuristics can look ahead arbitrarily-many levels in the search tree instead of a ‘shallow’ localised lookahead of classical heuristics. A restart-based search strategy allows for multiple learned models to be acquired and exploited in the solver’s optimisation. We demonstrate deep variable ordering heuristics based on the smallest, anti first-fail, and maximum regret heuristics. Results on instances from the MiniZinc benchmark suite show that deep heuristics solve 20% more problem instances while improving on overall runtime for the Open Stacks and Evilshop benchmark problems.
...
Solvers for constraint optimisation problems exploit variable and value ordering heuristics. Numerous expert-designed heuristics exist, while recent research learns novel, customised heuristics from past problem instances. This article addresses unseen problems for which no historical data is available. We propose one-shot learning of customised, problem instance-specific heuristics. To do so, we introduce the concept of deep heuristics, a data-driven approach to learn extended versions of a given variable ordering heuristic online. First, for a problem instance, an initial online probing phase collects data, from which a deep heuristic function is learned. The learned heuristics can look ahead arbitrarily-many levels in the search tree instead of a ‘shallow’ localised lookahead of classical heuristics. A restart-based search strategy allows for multiple learned models to be acquired and exploited in the solver’s optimisation. We demonstrate deep variable ordering heuristics based on the smallest, anti first-fail, and maximum regret heuristics. Results on instances from the MiniZinc benchmark suite show that deep heuristics solve 20% more problem instances while improving on overall runtime for the Open Stacks and Evilshop benchmark problems.
Journal article
(2025)
-
Ambrogio Maria Bernardelli, Stefano Gualandi, Simone Milanesi, Hoong Chuin Lau, Neil Yorke-Smith
Training neural networks (NNs) using combinatorial optimization solvers has gained attention in recent years. In low-data settings, the use of state-of-the-art mixed integer linear programming solvers, for instance, has the potential to exactly train an NN while avoiding computing-intensive training and hyperparameter tuning and simultaneously training and sparsifying the network. We study the case of few-bit discrete-valued neural networks, both binarized neural networks (BNNs) whose values are restricted to 61 and integer-valued neural networks (INNs) whose values lie in the range {―P, ::: , P}. Few-bit NNs receive increasing recognition because of their lightweight architecture and ability to run on low-power devices: for example, being implemented using Boolean operations. This paper proposes new methods to improve the training of BNNs and INNs. Our contribution is a multiobjective ensemble approach based on training a single NN for each possible pair of classes and applying a majority voting scheme to predict the final output. Our approach results in the training of robust sparsified networks whose output is not affected by small perturbations on the input and whose number of active weights is as small as possible. We empirically compare this BeMi approach with the current state of the art in solver-based NN training and with traditional gradient-based training, focusing on BNN learning in few-shot contexts. We compare the benefits and drawbacks of INNs versus BNNs, bringing new light to the distribution of weights over the {―P, ::: , P} interval. Finally, we compare multiobjective versus single-objective training of INNs, showing that robustness and network simplicity can be acquired simultaneously, thus obtaining better test performances. Although the previous state-of-the-art approaches achieve an average accuracy of 51:1% on the Modified National Institute of Standards and Technology data set, the BeMi ensemble approach achieves an average accuracy of 68.4% when trained with 10 images per class and 81.8% when trained with 40 images per class while having up to 75.3% NN links removed.
...
Training neural networks (NNs) using combinatorial optimization solvers has gained attention in recent years. In low-data settings, the use of state-of-the-art mixed integer linear programming solvers, for instance, has the potential to exactly train an NN while avoiding computing-intensive training and hyperparameter tuning and simultaneously training and sparsifying the network. We study the case of few-bit discrete-valued neural networks, both binarized neural networks (BNNs) whose values are restricted to 61 and integer-valued neural networks (INNs) whose values lie in the range {―P, ::: , P}. Few-bit NNs receive increasing recognition because of their lightweight architecture and ability to run on low-power devices: for example, being implemented using Boolean operations. This paper proposes new methods to improve the training of BNNs and INNs. Our contribution is a multiobjective ensemble approach based on training a single NN for each possible pair of classes and applying a majority voting scheme to predict the final output. Our approach results in the training of robust sparsified networks whose output is not affected by small perturbations on the input and whose number of active weights is as small as possible. We empirically compare this BeMi approach with the current state of the art in solver-based NN training and with traditional gradient-based training, focusing on BNN learning in few-shot contexts. We compare the benefits and drawbacks of INNs versus BNNs, bringing new light to the distribution of weights over the {―P, ::: , P} interval. Finally, we compare multiobjective versus single-objective training of INNs, showing that robustness and network simplicity can be acquired simultaneously, thus obtaining better test performances. Although the previous state-of-the-art approaches achieve an average accuracy of 51:1% on the Modified National Institute of Standards and Technology data set, the BeMi ensemble approach achieves an average accuracy of 68.4% when trained with 10 images per class and 81.8% when trained with 40 images per class while having up to 75.3% NN links removed.
The Dutch housing market comprises three sectors: social-rented, private-rented, and owner-occupied. The contemporary market is marked by a shortage of supply and a large subsidised social sector. Waiting lists for social housing are growing, whereas households with incomes above the limit do not or cannot leave the social sector. Government policy and market regulations change frequently, not least for political reasons. In view of commonly recognised problems in the housing market, this article considers the ‘internal demand’ of those households that are dissatisfied with their current residence. We examine the effects of regulatory policy by means of an exploratory agent-based simulation. The results provide perspectives on how internal demand is impacted by regulations in a housing market that is suffering from a shortage, and allow decision makers to weigh the pros and cons of policy measures.
...
The Dutch housing market comprises three sectors: social-rented, private-rented, and owner-occupied. The contemporary market is marked by a shortage of supply and a large subsidised social sector. Waiting lists for social housing are growing, whereas households with incomes above the limit do not or cannot leave the social sector. Government policy and market regulations change frequently, not least for political reasons. In view of commonly recognised problems in the housing market, this article considers the ‘internal demand’ of those households that are dissatisfied with their current residence. We examine the effects of regulatory policy by means of an exploratory agent-based simulation. The results provide perspectives on how internal demand is impacted by regulations in a housing market that is suffering from a shortage, and allow decision makers to weigh the pros and cons of policy measures.
Metaheuristics are known to be effective in finding good solutions in combinatorial optimization, but solving stochastic problems is costly due to the need for evaluation of multiple scenarios. We propose a general method to reduce the number of scenario evaluations per solution and thus improve metaheuristic efficiency. We use a sequential sampling procedure exploiting estimates of the solutions’ expected objective values. These values are obtained with a predictive model, which is founded on an estimated discrete probability distribution linearly related to all solutions’ objective distributions; the probability distribution is continuously refined based on incoming solution evaluation. The proposed method is tested using simulated annealing, but in general applicable to single solution metaheuristics. The method’s performance is compared to descriptive sampling and an adaptation of a sequential sampling method assuming noisy evaluations. Experimental results on three problems indicate the proposed method is robust overall, and performs better on average than the baselines on two of the problems.
...
Metaheuristics are known to be effective in finding good solutions in combinatorial optimization, but solving stochastic problems is costly due to the need for evaluation of multiple scenarios. We propose a general method to reduce the number of scenario evaluations per solution and thus improve metaheuristic efficiency. We use a sequential sampling procedure exploiting estimates of the solutions’ expected objective values. These values are obtained with a predictive model, which is founded on an estimated discrete probability distribution linearly related to all solutions’ objective distributions; the probability distribution is continuously refined based on incoming solution evaluation. The proposed method is tested using simulated annealing, but in general applicable to single solution metaheuristics. The method’s performance is compared to descriptive sampling and an adaptation of a sequential sampling method assuming noisy evaluations. Experimental results on three problems indicate the proposed method is robust overall, and performs better on average than the baselines on two of the problems.
Accurate predictions of power fluctuations are pivotal to the operation of flexibility markets. While the design of flexibility markets is an active and ongoing field of research, the question of how to elicit high quality predictions in a non-cooperative setting is often overlooked. Conceptually, we contribute the concept of best prediction incentivizing contracts. Under such contracts the best response of an agent is to report the true distribution of its power fluctuation. This concept differs from Incentive Compatibility by explicitly taking epistemic uncertainty into account: while Incentive Compatible mechanisms often assume the agent possess perfect knowledge of their own valuation, our concept incentivizes agents to reduce their epistemic uncertainty about the world. In practical terms, we present generic closed form solutions for polynomial distributions and show they can be used to approximate realistic Gaussian distributions. Lastly, placing our work in a larger context, we show that third party agents can profit from providing improved predictions via arbitrage.
...
Accurate predictions of power fluctuations are pivotal to the operation of flexibility markets. While the design of flexibility markets is an active and ongoing field of research, the question of how to elicit high quality predictions in a non-cooperative setting is often overlooked. Conceptually, we contribute the concept of best prediction incentivizing contracts. Under such contracts the best response of an agent is to report the true distribution of its power fluctuation. This concept differs from Incentive Compatibility by explicitly taking epistemic uncertainty into account: while Incentive Compatible mechanisms often assume the agent possess perfect knowledge of their own valuation, our concept incentivizes agents to reduce their epistemic uncertainty about the world. In practical terms, we present generic closed form solutions for polynomial distributions and show they can be used to approximate realistic Gaussian distributions. Lastly, placing our work in a larger context, we show that third party agents can profit from providing improved predictions via arbitrage.
Exploration in reinforcement learning remains a difficult challenge. In order to drive exploration, ensembles with randomized prior functions have recently been popularized to quantify uncertainty in the value model. There is no theoretical reason for these ensembles to resemble the actual posterior, however. In this work, we view training ensembles from the perspective of Sequential Monte Carlo, a Monte Carlo method that approximates a sequence of distributions with a set of particles. In particular, we propose an algorithm that exploits both the practical flexibility of ensembles and theory of the Bayesian paradigm. We incorporate this method into a standard Deep Q-learning agent (DQN) and experimentally show qualitatively good uncertainty quantification and improved exploration capabilities over a regular ensemble.
...
Exploration in reinforcement learning remains a difficult challenge. In order to drive exploration, ensembles with randomized prior functions have recently been popularized to quantify uncertainty in the value model. There is no theoretical reason for these ensembles to resemble the actual posterior, however. In this work, we view training ensembles from the perspective of Sequential Monte Carlo, a Monte Carlo method that approximates a sequence of distributions with a set of particles. In particular, we propose an algorithm that exploits both the practical flexibility of ensembles and theory of the Bayesian paradigm. We incorporate this method into a standard Deep Q-learning agent (DQN) and experimentally show qualitatively good uncertainty quantification and improved exploration capabilities over a regular ensemble.
Delftse Foundations of Computation is a textbook for a one quarter introductory course in theoretical computer science. It includes topics from propositional and predicate logic, proof techniques, set theory and the theory of computation, along with practical applications to computer science. It has no prerequisites other than a general familiarity with computer programming.
...
Delftse Foundations of Computation is a textbook for a one quarter introductory course in theoretical computer science. It includes topics from propositional and predicate logic, proof techniques, set theory and the theory of computation, along with practical applications to computer science. It has no prerequisites other than a general familiarity with computer programming.