Mark C. W. van Rossum
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Mark C. W. van Rossum has a total of 152 co-authors. Here are the top co-authors by number of shared publications:
- Theodorus Maria Nieuwenhuizen
Institute for Theoretical Physics, University of Amsterdam, Amsterdam, The Netherlands
8 shared publications - Sander Wessel Keemink
Affiliation Department of Machine Learning and Neural Computing, Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, Netherlands
7 shared publications - Matthew F. Nolan
Centre for Discovery Brain Sciences, University of Edinburgh
5 shared publications - Per Jesper Sjostrom
Centre for Research in Neuroscience, Brain Repair and Integrative Neuroscience Program, Department of Medicine, Department of Neurology and Neurosurgery, The Research Institute of the McGill University Health Centre, Montreal General Hospital
5 shared publications - Rui Ponte Costa
Centre for Neural Circuits and Behaviour, Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK
5 shared publications Aaron Pache
School of Mathematical Sciences, University of Nottingham, Nottingham NG7 2RD, United Kingdom
4 shared publications
Latest Publications
- Open Access
Mistake gating leads to energy and memory efficient continual learning
Synaptic plasticity is metabolically expensive, yet animals continuously update their internal models without exhausting energy reserves. However, when artificial neural networks are trained, the network parameters are typically updated on every sample that is presented, even if the sample was classified correctly. Inspired by the human negativity bias and error-related negativity, we propose ‘memorized mistake-gated learning’—a biologically plausible plasticity rule where synaptic updates are strictly gated by current and past classification errors. This reduces the number of updates the network needs to make by 50% ∼ 80%. Mistake gating is particularly well suited in two cases: 1) For incremental learning where new knowledge is acquired on a background of pre-existing knowledge, 2) For online learning scenarios when data needs to be stored for later replay, as mistake-gating reduces storage buffer requirements. The algorithm can be implemented in a few lines of code, adds no hyper-parameters, and comes at negligible computational overhead. Learning on mistakes is an energy efficient and biologically relevant modification to commonly used learning rules that is well suited for continual learning.
Open AccessMistake gating leads to energy and memory efficient continual learning
Synaptic plasticity is metabolically expensive, yet animals continuously update their internal models without exhausting energy reserves. However, when artificial neural networks are trained, the network parameters are typically updated on every sample that is presented, even if the sample was classified correctly. Inspired by the human negativity bias and error-related negativity, we propose ‘memorized mistake-gated learning’—a biologically plausible plasticity rule where synaptic updates are strictly gated by current and past classification errors. This reduces the number of updates the network needs to make by 50% ∼ 80%. Mistake gating is particularly well suited in two cases: 1) For incremental learning where new knowledge is acquired on a background of pre-existing knowledge, 2) For online learning scenarios when data needs to be stored for later replay, as mistake-gating reduces storage buffer requirements. The algorithm can be implemented in a few lines of code, adds no hyper-parameters, and comes at negligible computational overhead. Learning on mistakes is an energy efficient and biologically relevant modification to commonly used learning rules that is well suited for continual learning.
- Open Access
- 21 August 2025
- Published by eLife Sciences Publications, Ltd in eLife
- Vol. 14
Correction: Unified pre- and postsynaptic long-term plasticity enables reliable and flexible learning
- 21 August 2025
- Published by eLife Sciences Publications, Ltd in eLife
- Vol. 14
Open AccessCorrection: Unified pre- and postsynaptic long-term plasticity enables reliable and flexible learning
- Open Access
- 11 April 2025
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 21 (4) , e1012969
Biases in neural population codes with a few active neurons
Throughout the brain information is coded in the activity of multiple neurons at once, so called population codes. Population codes are a robust and accurate way of coding information. One can evaluate the quality of population coding by trying to read out the code with a decoder, and estimate the encoded stimulus. In particular when neurons are noisy, coding accuracy has extensively been evaluated in terms of the trial-to-trial variation in the estimate. While most decoders yield unbiased estimators if many neurons are actived, when only a few neurons are active, biases readily emerge. That is, even after averaging, a systematic difference between the true stimulus and its estimate remains. We characterize the shape of this bias for different encoding models (rectified cosine tuning and von Mises functions) and show that it can be both attractive or repulsive for different stimulus values. Biases appear for maximum likelihood and Bayesian decoders. The biases have a non-trivial dependence on noise. We also introduce a technique to estimate the bias and variance of Bayesian least square decoders. The work is of interest to those studying neural populations with a few active neurons.
- 11 April 2025
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 21 (4) , e1012969
Open AccessBiases in neural population codes with a few active neurons
Throughout the brain information is coded in the activity of multiple neurons at once, so called population codes. Population codes are a robust and accurate way of coding information. One can evaluate the quality of population coding by trying to read out the code with a decoder, and estimate the encoded stimulus. In particular when neurons are noisy, coding accuracy has extensively been evaluated in terms of the trial-to-trial variation in the estimate. While most decoders yield unbiased estimators if many neurons are actived, when only a few neurons are active, biases readily emerge. That is, even after averaging, a systematic difference between the true stimulus and its estimate remains. We characterize the shape of this bias for different encoding models (rectified cosine tuning and von Mises functions) and show that it can be both attractive or repulsive for different stimulus values. Biases appear for maximum likelihood and Bayesian decoders. The biases have a non-trivial dependence on noise. We also introduce a technique to estimate the bias and variance of Bayesian least square decoders. The work is of interest to those studying neural populations with a few active neurons.
- Open Access
- 28 October 2024
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 20 (10) , e1012554
Reinforcement learning when your life depends on it: A neuro-economic theory of learning
Synaptic plasticity enables animals to adapt to their environment, but memory formation can require a substantial amount of metabolic energy, potentially impairing survival. Hence, a neuro-economic dilemma arises whether learning is a profitable investment or not, and the brain must therefore judiciously regulate learning. Indeed, in experiments it was observed that during starvation, Drosophila suppress formation of energy-intensive aversive memories. Here we include energy considerations in a reinforcement learning framework. Simulated flies learned to avoid noxious stimuli through synaptic plasticity in either the energy expensive long-term memory (LTM) pathway, or the decaying anesthesia-resistant memory (ARM) pathway. The objective of the flies is to maximize their lifespan, which is calculated with a hazard function. We find that strategies that switch between the LTM and ARM pathways, based on energy reserve and reward prediction error, prolong lifespan. Our study highlights the significance of energy-regulation of memory pathways and dopaminergic control for adaptive learning and survival. It might also benefit engineering applications of reinforcement learning under resources constraints.
- 28 October 2024
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 20 (10) , e1012554
Open AccessReinforcement learning when your life depends on it: A neuro-economic theory of learning
Synaptic plasticity enables animals to adapt to their environment, but memory formation can require a substantial amount of metabolic energy, potentially impairing survival. Hence, a neuro-economic dilemma arises whether learning is a profitable investment or not, and the brain must therefore judiciously regulate learning. Indeed, in experiments it was observed that during starvation, Drosophila suppress formation of energy-intensive aversive memories. Here we include energy considerations in a reinforcement learning framework. Simulated flies learned to avoid noxious stimuli through synaptic plasticity in either the energy expensive long-term memory (LTM) pathway, or the decaying anesthesia-resistant memory (ARM) pathway. The objective of the flies is to maximize their lifespan, which is calculated with a hazard function. We find that strategies that switch between the LTM and ARM pathways, based on energy reserve and reward prediction error, prolong lifespan. Our study highlights the significance of energy-regulation of memory pathways and dopaminergic control for adaptive learning and survival. It might also benefit engineering applications of reinforcement learning under resources constraints.
- Open Access
- 28 October 2024
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 20 (10) , e1012553
Competitive plasticity to reduce the energetic costs of learning
The brain is not only constrained by energy needed to fuel computation, but it is also constrained by energy needed to form memories. Experiments have shown that learning simple conditioning tasks which might require only a few synaptic updates, already carries a significant metabolic cost. Yet, learning a task like MNIST to 95% accuracy appears to require at least 108 synaptic updates. Therefore the brain has likely evolved to be able to learn using as little energy as possible. We explored the energy required for learning in feedforward neural networks. Based on a parsimonious energy model, we propose two plasticity restricting algorithms that save energy: 1) only modify synapses with large updates, and 2) restrict plasticity to subsets of synapses that form a path through the network. In biology networks are often much larger than the task requires, yet vanilla backprop prescribes to update all synapses. In particular in this case, large savings can be achieved while only incurring a slightly worse learning time. Thus competitively restricting plasticity helps to save metabolic energy associated to synaptic plasticity. The results might lead to a better understanding of biological plasticity and a better match between artificial and biological learning. Moreover, the algorithms might benefit hardware because also electronic memory storage is energetically costly.
- 28 October 2024
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 20 (10) , e1012553
Open AccessCompetitive plasticity to reduce the energetic costs of learning
The brain is not only constrained by energy needed to fuel computation, but it is also constrained by energy needed to form memories. Experiments have shown that learning simple conditioning tasks which might require only a few synaptic updates, already carries a significant metabolic cost. Yet, learning a task like MNIST to 95% accuracy appears to require at least 108 synaptic updates. Therefore the brain has likely evolved to be able to learn using as little energy as possible. We explored the energy required for learning in feedforward neural networks. Based on a parsimonious energy model, we propose two plasticity restricting algorithms that save energy: 1) only modify synapses with large updates, and 2) restrict plasticity to subsets of synapses that form a path through the network. In biology networks are often much larger than the task requires, yet vanilla backprop prescribes to update all synapses. In particular in this case, large savings can be achieved while only incurring a slightly worse learning time. Thus competitively restricting plasticity helps to save metabolic energy associated to synaptic plasticity. The results might lead to a better understanding of biological plasticity and a better match between artificial and biological learning. Moreover, the algorithms might benefit hardware because also electronic memory storage is energetically costly.
- Open Access
Biases in population codes with a few active neurons
Throughout the brain information is coded in the activity of multiple neurons at once, so called population codes. Population codes are a robust and accurate way of coding information. One can evaluate the quality of population coding by trying to read out the code with a decoder, and estimate the encoded stimulus. Coding quality has traditionally been evaluated in terms of the trial-to-trial variation in the estimate. However, codes can also display biases. While most decoders yield unbiased estimators in the limit of many active neurons, we find that when only few neurons are active, biases readily emerge for many decoders. We show that the biases turn out to have a non-trivial dependence on noise and tuning curve properties. We also introduce a technique to estimate the bias and variance of Bayesian decoders. Overall, the work expands our understanding of population coding.
Open AccessBiases in population codes with a few active neurons
Throughout the brain information is coded in the activity of multiple neurons at once, so called population codes. Population codes are a robust and accurate way of coding information. One can evaluate the quality of population coding by trying to read out the code with a decoder, and estimate the encoded stimulus. Coding quality has traditionally been evaluated in terms of the trial-to-trial variation in the estimate. However, codes can also display biases. While most decoders yield unbiased estimators in the limit of many active neurons, we find that when only few neurons are active, biases readily emerge for many decoders. We show that the biases turn out to have a non-trivial dependence on noise and tuning curve properties. We also introduce a technique to estimate the bias and variance of Bayesian decoders. Overall, the work expands our understanding of population coding.
- Open Access
Reinforcement learning when your life depends on it: a neuro-economic theory of learning
Synaptic plasticity enables animals to adapt to their environment, but memory formation can consume a substantial amount of metabolic energy, potentially impairing survival. Hence, a neuro-economic dilemma arises whether learning is a profitable investment or not, and the brain must therefore judiciously regulate learning. Indeed, in experiments it was observed that during starvation, Drosophila suppress formation of energy-intensive aversive memories. Here we include energy considerations in a reinforcement learning framework. Simulated flies learned to avoid noxious stimuli through synaptic plasticity in either the energy expensive long-term memory (LTM) pathway, or the decaying anesthesia-resistant memory (ARM) pathway. The objective of the flies is to maximize their lifespan, which is calculated with a hazard function. We find that strategies that switch between the LTM and ARM pathways based on energy reserve and reward prediction error, prolong lifespan. Our study highlights the significance of energy-regulation of memory pathways and dopaminergic control for adaptive learning and survival. It might also benefit engineering applications of reinforcement learning under resources constraints.
Open AccessReinforcement learning when your life depends on it: a neuro-economic theory of learning
Synaptic plasticity enables animals to adapt to their environment, but memory formation can consume a substantial amount of metabolic energy, potentially impairing survival. Hence, a neuro-economic dilemma arises whether learning is a profitable investment or not, and the brain must therefore judiciously regulate learning. Indeed, in experiments it was observed that during starvation, Drosophila suppress formation of energy-intensive aversive memories. Here we include energy considerations in a reinforcement learning framework. Simulated flies learned to avoid noxious stimuli through synaptic plasticity in either the energy expensive long-term memory (LTM) pathway, or the decaying anesthesia-resistant memory (ARM) pathway. The objective of the flies is to maximize their lifespan, which is calculated with a hazard function. We find that strategies that switch between the LTM and ARM pathways based on energy reserve and reward prediction error, prolong lifespan. Our study highlights the significance of energy-regulation of memory pathways and dopaminergic control for adaptive learning and survival. It might also benefit engineering applications of reinforcement learning under resources constraints.
- Open Access
- 4 September 2023
- Published by Elsevier in Current Opinion in Neurobiology
- Vol. 83, 102779
Energetically efficient learning in neuronal networks
- 4 September 2023
- Published by Elsevier in Current Opinion in Neurobiology
- Vol. 83, 102779
Open AccessEnergetically efficient learning in neuronal networks
- Open Access
Competitive plasticity to reduce the energetic costs of learning
The brain is not only constrained by energy needed to fuel computation, but it is also constrained by energy needed to form memories. Experiments have shown that learning simple conditioning tasks already carries a significant metabolic cost. Yet, learning a task like MNIST to 95% accuracy appears to require at least 10 8 synaptic updates. Therefore the brain has likely evolved to be able to learn using as little energy as possible. We explored the energy required for learning in feedforward neural networks. Based on a parsimonious energy model, we propose two plasticity restricting algorithms that save energy: 1) only modify synapses with large updates, and 2) restrict plasticity to subsets of synapses that form a path through the network. Combining these two methods leads to substantial energy savings while only incurring a small increase in learning time. In biology networks are often much larger than the task requires. In particular in that case, large savings can be achieved. Thus competitively restricting plasticity helps to save metabolic energy associated to synaptic plasticity. The results might lead to a better understanding of biological plasticity and a better match between artificial and biological learning. Moreover, the algorithms might also benefit hardware because in electronics memory storage is energetically costly as well.
Open AccessCompetitive plasticity to reduce the energetic costs of learning
The brain is not only constrained by energy needed to fuel computation, but it is also constrained by energy needed to form memories. Experiments have shown that learning simple conditioning tasks already carries a significant metabolic cost. Yet, learning a task like MNIST to 95% accuracy appears to require at least 10 8 synaptic updates. Therefore the brain has likely evolved to be able to learn using as little energy as possible. We explored the energy required for learning in feedforward neural networks. Based on a parsimonious energy model, we propose two plasticity restricting algorithms that save energy: 1) only modify synapses with large updates, and 2) restrict plasticity to subsets of synapses that form a path through the network. Combining these two methods leads to substantial energy savings while only incurring a small increase in learning time. In biology networks are often much larger than the task requires. In particular in that case, large savings can be achieved. Thus competitively restricting plasticity helps to save metabolic energy associated to synaptic plasticity. The results might lead to a better understanding of biological plasticity and a better match between artificial and biological learning. Moreover, the algorithms might also benefit hardware because in electronics memory storage is energetically costly as well.
- Open Access
- 28 March 2023
- Published by Society for Neuroscience in Journal of Neuroscience
- Vol. 43 (21) , 3838-3848
Rule Abstraction Is Facilitated by Auditory Cuing in REM Sleep
Sleep facilitates abstraction, but the exact mechanisms underpinning this are unknown. Here, we aimed to determine whether triggering reactivation in sleep could facilitate this process. We paired abstraction problems with sounds, then replayed these during either slow-wave sleep (SWS) or rapid eye movement (REM) sleep to trigger memory reactivation in 27 human participants (19 female). This revealed performance improvements on abstraction problems that were cued in REM, but not problems cued in SWS. Interestingly, the cue-related improvement was not significant until a follow-up retest 1 week after the manipulation, suggesting that REM may initiate a sequence of plasticity events that requires more time to be implemented. Furthermore, memory-linked trigger sounds evoked distinct neural responses in REM, but not SWS. Overall, our findings suggest that targeted memory reactivation in REM can facilitate visual rule abstraction, although this effect takes time to unfold.SIGNIFICANCE STATEMENTThe ability to abstract rules from a corpus of experiences is a building block of human reasoning. Sleep is known to facilitate rule abstraction, but it remains unclear whether we can manipulate this process actively and which stage of sleep is most important. Targeted memory reactivation (TMR) is a technique that uses re-exposure to learning-related sensory cues during sleep to enhance memory consolidation. Here, we show that TMR, when applied during REM sleep, can facilitate the complex recombining of information needed for rule abstraction. Furthermore, we show that this qualitative REM-related benefit emerges over the course of a week after learning, suggesting that memory integration may require a slower form of plasticity.
- 28 March 2023
- Published by Society for Neuroscience in Journal of Neuroscience
- Vol. 43 (21) , 3838-3848
Open AccessRule Abstraction Is Facilitated by Auditory Cuing in REM Sleep
Sleep facilitates abstraction, but the exact mechanisms underpinning this are unknown. Here, we aimed to determine whether triggering reactivation in sleep could facilitate this process. We paired abstraction problems with sounds, then replayed these during either slow-wave sleep (SWS) or rapid eye movement (REM) sleep to trigger memory reactivation in 27 human participants (19 female). This revealed performance improvements on abstraction problems that were cued in REM, but not problems cued in SWS. Interestingly, the cue-related improvement was not significant until a follow-up retest 1 week after the manipulation, suggesting that REM may initiate a sequence of plasticity events that requires more time to be implemented. Furthermore, memory-linked trigger sounds evoked distinct neural responses in REM, but not SWS. Overall, our findings suggest that targeted memory reactivation in REM can facilitate visual rule abstraction, although this effect takes time to unfold.SIGNIFICANCE STATEMENTThe ability to abstract rules from a corpus of experiences is a building block of human reasoning. Sleep is known to facilitate rule abstraction, but it remains unclear whether we can manipulate this process actively and which stage of sleep is most important. Targeted memory reactivation (TMR) is a technique that uses re-exposure to learning-related sensory cues during sleep to enhance memory consolidation. Here, we show that TMR, when applied during REM sleep, can facilitate the complex recombining of information needed for rule abstraction. Furthermore, we show that this qualitative REM-related benefit emerges over the course of a week after learning, suggesting that memory integration may require a slower form of plasticity.