arXiv subjects

Nicholas Guttenberg

Publications and source records attributed to Nicholas Guttenberg.

At least 19 recordsLinked to original sources

arXiv2021

Bootstrapping of memetic from genetic evolution via inter-agent selection pressures

We create an artificial system of agents (attention-based neural networks) which selectively exchange messages with each-other in order to study the emergence of memetic evolution and how memetic evolutionary pressures interact with genetic evolution of the network weights. We observe that the ability of agents to exert selection pressures on each-other is essential for memetic evolution to bootstrap itself into a state which has both high-fidelity replication of memes, as well as continuing production of new memes over time. However, in this system there is very little interaction between this memetic 'ecology' and underlying tasks driving individual fitness - the emergent meme layer appears to be neither helpful nor harmful to agents' ability to learn to solve tasks. Sourcecode for these experiments is available at https://github.com/GoodAI/memes

cs.AI↗

arXiv2021

Evolutionary rates of information gain and decay in fluctuating environments

In this paper, we wish to investigate the dynamics of information transfer in evolutionary dynamics. We use information theoretic tools to track how much information an evolving population has obtained and managed to retain about different environments that it is exposed to. By understanding the dynamics of information gain and loss in a static environment, we predict how that same evolutionary system would behave when the environment is fluctuating. Specifically, we anticipate a cross-over between the regime in which fluctuations improve the ability of the evolutionary system to capture environmental information and the regime in which the fluctuations inhibit it, governed by a cross-over in the timescales of information gain and decay.

q-bio.PE↗

arXiv2020

Earth-Like: An education & outreach tool for exploring the diversity of planets like our own

Earth-Like is an interactive website and twitter bot that allows users to explore changes in the average global surface temperature of an Earth-like planet due to variations in the surface oceans and emerged land coverage, rate of volcanism (degassing), and the level of the received solar radiation. The temperature is calculated using a simple carbon-silicate cycle model to change the level of $\rm CO_2$ in the atmosphere based on the chosen parameters. The model can achieve a temperature range exceeding $-100^\circ$C to $100^\circ$C by varying all three parameters, including freeze-thaw cycles for a planet with our present-day volcanism rate and emerged land fraction situated at the outer edge of the habitable zone. To increase engagement, the planet is visualised by using a neural network to render an animated globe, based on the calculated average surface temperature and chosen values for land fraction and volcanism. The website and bot can be found at earthlike.world and on twitter as @earthlikeworld. Initial feedback via a user survey suggested that Earth-Like is effective at demonstrating that minor changes in planetary properties can strongly impact the surface environment. The goal of the project is to increase understanding of the challenges we face in finding another habitable planet due to the likely diversity of conditions on rocky worlds within our Galaxy.

astro-ph.EP↗

arXiv2019

BADGER: Learning to (Learn [Learning Algorithms] through Multi-Agent Communication)

In this work, we propose a novel memory-based multi-agent meta-learning architecture and learning procedure that allows for learning of a shared communication policy that enables the emergence of rapid adaptation to new and unseen environments by learning to learn learning algorithms through communication. Behavior, adaptation and learning to adapt emerges from the interactions of homogeneous experts inside a single agent. The proposed architecture should allow for generalization beyond the level seen in existing methods, in part due to the use of a single policy shared by all experts within the agent as well as the inherent modularity of 'Badger'.

cs.AI↗

arXiv2019

Estimating Planetary Mass with Deep Learning

While thousands of exoplanets have been confirmed, the known properties about individual discoveries remain sparse and depend on detection technique. To utilize more than a small section of the exoplanet dataset, tools need to be developed to estimate missing values based on the known measurements. Here, we demonstrate the use of a neural network that models the density of planets in a space of six properties that is then used to impute a probability distribution for missing values. Our results focus on planetary mass which neither the radial velocity nor transit techniques for planet identification can provide alone. The neural network can impute mass across the four orders of magnitude in the exoplanet archive, and return a distribution of masses for each planet that can inform about trends in the underlying dataset. The average error on this mass estimate from a radial velocity detection is a factor of 1.5 of the observed value, and 2.7 for a transit observation. The mass of Proxima Centauri b found by this method is $1.6^{\rm +0.46}_{\rm -0.36}$ M$_\oplus$, where the upper and lower bounds are derived from the root mean square deviation from the log mass probability distribution. The network can similarly impute the other potentially missing properties, and we use this to predict planet radius for radial velocity measurements, with an average error of a factor 1.4 of the observed value. The ability of neural networks to search for patterns in multidimensional data means that such techniques have the potential to greatly expand the use of the exoplanet catalogue.

astro-ph.EP↗

arXiv2019

Deep learning on butterfly phenotypes tests evolution's oldest mathematical model

Traditional anatomical analyses captured only a fraction of real phenomic information. Here, we apply deep learning to quantify total phenotypic similarity across 2468 butterfly photographs, covering 38 subspecies from the polymorphic mimicry complex of $\textit{Heliconius erato}$ and $\textit{Heliconius melpomene}$. Euclidean phenotypic distances, calculated using a deep convolutional triplet network, demonstrate significant convergence between interspecies co-mimics. This quantitatively validates a key prediction of Müllerian mimicry theory, evolutionary biology's oldest mathematical model. Phenotypic neighbor-joining trees are significantly correlated with wing pattern gene phylogenies, demonstrating objective, phylogenetically informative phenome capture. Comparative analyses indicate frequency-dependent, mutual convergence with coevolutionary exchange of wing pattern features. Therefore, phenotypic analysis supports reciprocal coevolution, predicted by classical mimicry theory but since disputed, and reveals mutual convergence as an intrinsic generator for the surprising diversity of Müllerian mimicry. This demonstrates that deep learning can generate phenomic spatial embeddings which enable quantitative tests of evolutionary hypotheses previously only testable subjectively.

q-bio.PE↗

arXiv2019

Generating the support with extreme value losses

When optimizing against the mean loss over a distribution of predictions in the context of a regression task, then even if there is a distribution of targets the optimal prediction distribution is always a delta function at a single value. Methods of constructing generative models need to overcome this tendency. We consider a simple method of summarizing the prediction error, such that the optimal strategy corresponds to outputting a distribution of predictions with a support that matches the support of the distribution of targets --- optimizing against the minimal value of the loss given a set of samples from the prediction distribution, rather than the mean. We show that models trained against this loss learn to capture the support of the target distribution and, when combined with an auxiliary classifier-like prediction task, can be projected via rejection sampling to reproduce the full distribution of targets. The resulting method works well compared to other generative modeling approaches particularly in low dimensional spaces with highly non-trivial distributions, due to mode collapse solutions being globally suboptimal with respect to the extreme value loss. However, the method is less suited to high-dimensional spaces such as images due to the scaling of the number of samples needed in order to accurately estimate the extreme value loss when the dimension of the data manifold becomes large.

cs.LG↗

arXiv2018

On the potential for open-endedness in neural networks

Natural evolution gives the impression of leading to an open-ended process of increasing diversity and complexity. If our goal is to produce such open-endedness artificially, this suggests an approach driven by evolutionary metaphor. On the other hand, techniques from machine learning and artificial intelligence are often considered too narrow to provide the sort of exploratory dynamics associated with evolution. In this paper, we hope to bridge that gap by reviewing common barriers to open-endedness in the evolution-inspired approach and how they are dealt with in the evolutionary case - collapse of diversity, saturation of complexity, and failure to form new kinds of individuality. We then show how these problems map onto similar issues in the machine learning approach, and discuss how the same insights and solutions which alleviated those barriers in evolutionary approaches can be ported over. At the same time, the form these issues take in the machine learning formulation suggests new ways to analyze and resolve barriers to open-endedness. Ultimately, we hope to inspire researchers to be able to interchangeably use evolutionary and gradient-descent-based machine learning methods to approach the design and creation of open-ended systems.

nlin.AO↗

arXiv2018

Being curious about the answers to questions: novelty search with learned attention

We investigate the use of attentional neural network layers in order to learn a `behavior characterization' which can be used to drive novelty search and curiosity-based policies. The space is structured towards answering a particular distribution of questions, which are used in a supervised way to train the attentional neural network. We find that in a 2d exploration task, the structure of the space successfully encodes local sensory-motor contingencies such that even a greedy local `do the most novel action' policy with no reinforcement learning or evolution can explore the space quickly. We also apply this to a high/low number guessing game task, and find that guessing according to the learned attention profile performs active inference and can discover the correct number more quickly than an exact but passive approach.

cs.AI↗

arXiv2018

Hydrodynamic Theory of Flocking in the Presence of Quenched Disorder

The effect of quenched (frozen) orientational disorder on the collective motion of active particles is analyzed. We find that, as with annealed disorder (Langevin noise), active polar systems are far more robust against quenched disorder than their equilibrium counterparts. In particular, long ranged order (i.e., the existence of a non-zero average velocity $\langle {\bf v} \rangle$) persists in the presence of quenched disorder even in spatial dimensions $d=3$, while it is destroyed even by arbitrarily weak disorder in $d \le 4$ in equilibrium systems. Furthermore, in $d=2$, quasi-long-ranged order (i.e., spatial velocity correlations that decay as a power law with distance) occurs when quenched disorder is present, in contrast to the short-ranged order that is all that can survive in equilibrium. These predictions are borne out by simulations in both two and three dimensions.

cond-mat.stat-mech↗

arXiv2018

Swarming in the Dirt: Ordered Flocks with Quenched Disorder

The effect of quenched (frozen) disorder on the collective motion of active particles is analyzed. We find that active polar systems are far more robust against quenched disorder than equilibrium ferromagnets. Long ranged order (a non-zero average velocity $\langle{\bf v}\rangle$) persists in the presence of quenched disorder even in spatial dimensions $d=3$; in $d=2$, quasi-long-ranged order (i.e., spatial velocity correlations that decay as a power law with distance) occurs. In equilibrium systems, only quasi-long-ranged order in $d=3$ and short ranged order in $d=2$ are possible. Our theoretical predictions for two dimensions are borne out by simulations.

cond-mat.stat-mech↗

arXiv2018

Learning to generate classifiers

We train a network to generate mappings between training sets and classification policies (a 'classifier generator') by conditioning on the entire training set via an attentional mechanism. The network is directly optimized for test set performance on an training set of related tasks, which is then transferred to unseen 'test' tasks. We use this to optimize for performance in the low-data and unsupervised learning regimes, and obtain significantly better performance in the 10-50 datapoint regime than support vector classifiers, random forests, XGBoost, and k-nearest neighbors on a range of small datasets.

cs.LG↗

arXiv2017

Learning body-affordances to simplify action spaces

Controlling embodied agents with many actuated degrees of freedom is a challenging task. We propose a method that can discover and interpolate between context dependent high-level actions or body-affordances. These provide an abstract, low-dimensional interface indexing high-dimensional and time- extended action policies. Our method is related to recent ap- proaches in the machine learning literature but is conceptually simpler and easier to implement. More specifically our method requires the choice of a n-dimensional target sensor space that is endowed with a distance metric. The method then learns an also n-dimensional embedding of possibly reactive body-affordances that spread as far as possible throughout the target sensor space.

cs.AI↗

arXiv2017

Selection first path to the origin of life

We propose an alternative to the prevailing two origin of life narratives, one based on a replicator first hypothesis, and one based on a metabolism first hypothesis. Both hypotheses have known difficulties: All known evolvable molecular replicators such as RNA require complex chemical (enzymatic) machinery for the replication process. Likewise, contemporary cellular metabolisms require several enzymatically catalyzed steps, and it is difficult to identify a non-enzymatic path to their realization. We propose that there must have been precursors to both replication and metabolism that enable a form of selection to take place through action of simple chemical and physical processes. We model a concrete example of such a process, repeated sequestration of binary molecular combinations after exposure to an environment with a broad distribution of chemical components, as might be realized experimentally in in a repeated wet-dry cycle. We show that the repeated sequestration dynamics results in a selective amplification of a very small subset of molecular species present in the environment, thus providing a candidate primordial selection process.

q-bio.PE↗

arXiv2017

Counterfactual Control for Free from Generative Models

We introduce a method by which a generative model learning the joint distribution between actions and future states can be used to automatically infer a control scheme for any desired reward function, which may be altered on the fly without retraining the model. In this method, the problem of action selection is reduced to one of gradient descent on the latent space of the generative model, with the model itself providing the means of evaluating outcomes and finding the gradient, much like how the reward network in Deep Q-Networks (DQN) provides gradient information for the action generator. Unlike DQN or Actor-Critic, which are conditional models for a specific reward, using a generative model of the full joint distribution permits the reward to be changed on the fly. In addition, the generated futures can be inspected to gain insight in to what the network 'thinks' will happen, and to what went wrong when the outcomes deviate from prediction.

cs.LG↗

arXiv2016

Permutation-equivariant neural networks applied to dynamics prediction

The introduction of convolutional layers greatly advanced the performance of neural networks on image tasks due to innately capturing a way of encoding and learning translation-invariant operations, matching one of the underlying symmetries of the image domain. In comparison, there are a number of problems in which there are a number of different inputs which are all 'of the same type' --- multiple particles, multiple agents, multiple stock prices, etc. The corresponding symmetry to this is permutation symmetry, in that the algorithm should not depend on the specific ordering of the input data. We discuss a permutation-invariant neural network layer in analogy to convolutional layers, and show the ability of this architecture to learn to predict the motion of a variable number of interacting hard discs in 2D. In the same way that convolutional layers can generalize to different image sizes, the permutation layer we describe generalizes to different numbers of objects.

cs.CV↗

arXiv2016

Neural Coarse-Graining: Extracting slowly-varying latent degrees of freedom with neural networks

We present a loss function for neural networks that encompasses an idea of trivial versus non-trivial predictions, such that the network jointly determines its own prediction goals and learns to satisfy them. This permits the network to choose sub-sets of a problem which are most amenable to its abilities to focus on solving, while discarding 'distracting' elements that interfere with its learning. To do this, the network first transforms the raw data into a higher-level categorical representation, and then trains a predictor from that new time series to its future. To prevent a trivial solution of mapping the signal to zero, we introduce a measure of non-triviality via a contrast between the prediction error of the learned model with a naive model of the overall signal statistics. The transform can learn to discard uninformative and unpredictable components of the signal in favor of the features which are both highly predictive and highly predictable. This creates a coarse-grained model of the time-series dynamics, focusing on predicting the slowly varying latent parameters which control the statistics of the time-series, rather than predicting the fast details directly. The result is a semi-supervised algorithm which is capable of extracting latent parameters, segmenting sections of time-series with differing statistics, and building a higher-level representation of the underlying dynamics from unlabeled data.

cs.AI↗

arXiv2015

Emergence of self-reinforcing information bottlenecks in multilevel selection

We explain how hierarchical organization of biological systems emerges naturally during evolution, through a transition in the units of individuality. We will show how these transitions are the result of competing selective forces operating at different levels of organization, each level having different units of individuality. Such a transition represents a singular point in the evolutionary process, which we will show corresponds to a phase transition in the way information is encoded, with the formation of self-reinforcing information bottlenecks. We present an abstract model for characterizing these transitions that is quite general, applicable to many different versions of such transitions. As a concrete example, we consider the transition to multicellularity. Specifically, we study a stochastic model where isolated communities of interacting individuals (e.g. cells) undergo a transition to higher-order individuality (e.g. multicellularity). This transition is indicated by the marked decrease in the number of cells utilized to generate new communities from pre-existing ones. In this sense, the community begins to reproduce as a whole via a decreasing number of cells. We show that the fitness barrier to this transition is strongly reduced by horizontal gene transfer. These features capture two of the most prominent aspects of the transition to multicellularity: the evolution of a developmental process and reproduction through a unicellular bottleneck.

q-bio.PE↗