Life, learning and emergence
Developing mathematical foundations for life and artificial life, machine intelligence, and other emergent phenomena that defy reductionism.
What is life? Darwin’s theory provides a qualitative understanding of evolution. But from a physics perspective, we don’t know how life got started in the first place. We investigate the thermodynamic basis for emergent self-replication and adaptation, of which biology is just one instance. Can this be used to engineer artificial digital life? Can evolution itself be made a predictive science?
How do we make intelligent machines? Far from approaching artificial general intelligence, AI is stuck in high-dimensional curve-fitting. We seek mathematical insights that could lead to more intelligent AI, such as causal reasoning, reusable functional modules, and a representation of the environment. We investigate ways to use computation and AI to automate the search for new mathematical insights. Are there fundamental limits to AI, and what might this tell us about human intelligence?
What are the emergent properties of digital and neural computation, and might this shed light on autonomy and free will? We study information processing at the genetic level and the functional architecture of gene regulatory networks. We seek a theoretical understanding of cell programming and how to infer programming sets. Is causality itself an emergent phenomenon, as we traverse across different organisational length scales?
LLMs generate texts that can fool us to credit them with human capabilities. Understanding their limitations is a key to using them wisely.
A new open-source platform is specifically tailored for developing complex dialogue systems, like generative conversational AI assistants.
Models trained on a Russian topical dataset, of knowledge-grounded human-human conversation, are capable of real-world tasks across languages.
A family of transformer-based DNA language models can interpret genomic sequences, opening new possibilities for complex biological research.
The quadratic complexity of attention in transformers is tackled by combining token-based memory and segment-level recurrence, using RMT.
The bipartite nature of regulatory networks means gene-gene logics are composed, which severely restricts which ones can show up in life.
Bursting cells can introduce noise in transcription factor screens, but modelling this process allows us to discern true counts from false.
Balancing memory from linear components with nonlinearities from memristors optimises the computational capacity of electronic reservoirs.
The eigenvalues of the mortality equation fall into two classes—the flower and the stem—but only the stem eigenvalues control the dynamics.
The structural and functional building blocks of gene regulatory networks correspond, which tell us how genetic computation is organised.
The underlying scale invariance properties of naturally occurring networks are often clouded by finite-size effects due to the sample data.
Scale-invariant plant clusters explain the ability for a diverse range of plant species to coexist in ecosystems such as Barra Colorado.
The notion of quantum superposition speeds up the training process for binary neural networks and ensures that their parameters are optimal.
A delicate balance between white blood cell protein expression and the molecules on the surface of tumour cells determines cancer prognoses.
Circuits of memristors, resistors with memory, can exhibit instabilities which allow classical tunnelling through potential energy barriers.
Fire sales of common asset holdings can whip through a channel of contagion between banks, insurance companies and investments funds.
Cancer patients who contract and recover from Coronavirus-2 exhibit long-term immune system weaknesses, depending on the type of cancer.
The mortality equation governs the dynamics of an evolving population with a given maximum age, offering a theory for programmed ageing.
Exact methods supersede approximations used in high-dimensional linear regression to find correlations in statistical physics problems.
The fraction of logics that are biologically permitted can be bounded and shown to be tiny, which makes inferring them from experiments easier.
The ability of deep neural networks to generalize can be unraveled using path integral methods to compute their typical Boolean functions.
Statistical methods that normally fail for very high-dimensional data can be rescued via mathematical tools from statistical physics.
We optimize Bayesian data clustering by mapping the problem to the statistical physics of a gas and calculating the lowest entropy state.
Machine learning techniques enhance the efficiency of energy harvesters by implementing reversible energy-conserving operations.
A simple solvable model of memristive networks suggests a correspondence between the asymptotic states of memristors and the Ising model.
Statistical physics harnesses links between maximum entropy and information theory to capture null model and real-world network features.
One-shot analogs of fluctuation-theorem results help unify these two approaches for small-scale, nonequilibrium statistical physics.
Network users who have access to the network’s most informative node, as quantified by a novel index, the InfoRank, have a competitive edge.
Exact solutions for the dynamics of interacting memristors predict whether they relax to higher or lower resistance states given random initialisations.
The distributions of size and shape of a material’s grains can be constructed from a 2D slice of the material and electron diffraction data.
A novel approach to volunteer clouds outperforms traditional distributed task scheduling algorithms in the presence of intensive workloads.
Bipartite networks model the structures of ecological and economic real-world systems, enabling hypothesis testing and crisis forecasting.
An iterative version of a method to identify hierarchies and rankings of nodes in directed networks can partly overcome its resolution limit.
The structure of two-dimensional borane, a new semi-metallic single-layered material, has two Dirac cones that meet right at the Fermi energy.
We generalise neural networks into a quantum framework, demonstrating the possibility of quantum auto-encoders and teleportation.
When people operate in echo chambers, they focus on information adhering to their system of beliefs. Debunking them is harder than it seems
Memristive networks preserve memory and have the ability to learn according to analysis of the network’s internal memory dynamics.
Exact equations of motion provide an analytical description of the evolution and relaxation properties of complex memristive circuits.
Moment-based methods provide a simple way to describe a population of spherical particles and extract 3d information from 2d measurements.
Inference from single snapshots of temporal networks can misleadingly group communities if the links between snapshots are correlated.
Spectroscopy experiments show that energy shifts due to photon emission from individual molecules satisfy a fundamental quantum relation.
A new equality which depends on the maximum entropy describes the worst-case amount of work done by finite-dimensional quantum systems.
The spectral density of graph ensembles provides an exact solution to the graph partitioning problem and helps detect community structure.
An extension of the Kelly criterion maximises the growth rate of multiplicative stochastic processes when limited resources are available.
With inspiration from Maxwell’s classic thought experiment, it is possible to extract macroscopic work from microscopic measurements of photons.
The principal eigenvalue of small neutral networks determines their robustness, and is bounded by the logarithm of the number of vertices.
An adaptive network of oscillators in fragmented and incoherent states can re-organise itself into connected and synchronized states.
Compact heat exchangers can be designed to run at low power if the exchange is concentrated in a crumpled surface fed by a fractal network.
The community matrix of a complex ecosystem captures the population dynamics of interacting species and transitions to unstable abundances.
The structural properties of a network motif predict its functional versatility and relate to gene regulatory networks.
Percolation theory shows that the formation of giant clusters of neurons relies on a few parameters that could be measured experimentally.
A subset of bootstrap percolation models, which stabilise systems of cells on infinite lattices, exhibit non-trivial phase transitions.
Properties of protein interaction networks test the reliability of data and hint at the underlying mechanism with which proteins recruit each other.
Single-shot information theory inspires a new formulation of statistical mechanics which measures the optimal guaranteed work of a system.
The stable structures of calcium and magnesium carbonate at high pressures are crucial for understanding the Earth's deep carbon cycle.
A local model of preferential attachment with short-term memory generates scale-free networks, which can be readily computed by memristors.
A simple formula gives the maximum time for an n x n grid to become entirely infected having undergone a bootstrap percolation process.
When networks come under attack, a repairable architecture is superior to, and globally distinct from, an architecture that is robust.
A review of the achievements concerning typical bipartite entanglement for random quantum states involving a large number of particles.
The critical probability for bootstrap percolation, a process which mimics the spread of an infection in a graph, is bounded for Galton-Watson trees.
Lognormal distributions (and mixtures of same) are a useful model for the size distribution in emulsions and sediments.
The immune system must simultaneously recall multiple defense strategies because many antigens can attack the host at the same time.
Information theory fixes weighted networks’ degeneracy issues with a generalisation of binary graphs and an optimal scale of link intensities.
Associative networks with different loads model the ability of the immune system to respond simultaneously to multiple distinct antigen invasions.
Information about 10% of the links in a complex network is sufficient to reconstruct its main features and resilience with the fitness model.
A statistical procedure identifies dominant edges within weighted networks to determine whether a network has reached its steady state.
A systematic way to vary the power-law scaling relations between loading parameters and volume of material aids the hierarchical design process.
Network analysis of diagnostic data identifies combinations of the key factors which cause Class III malocclusion and how they evolve over time.
Spectral analysis shows that disassortative networks exhibit a higher epidemiological threshold and are therefore easier to immunize.
Edge multiplicity—the number of triangles attached to edges—is a powerful analytic tool to understand and generalize network properties.
Methods from tailored random graph theory reveal the relation between true biological networks and the often-biased samples taken from them.
Analysis of the linear elastic behaviour of plant cell dispersions improves our understanding of how to stabilise and texturise food products.
A transfer operator formalism solves the macroscopic dynamics of disordered Ising chain systems which are relevant for ageing phenomena.
A Monte Carlo model simulates the microstructural evolution of metallic and ceramic powders during the consolidation process liquid-phase sintering.
The information needed to self-assemble a structure quantifies its modularity and explains the prevalence of certain structures over others.
Of the 256 elementary cellular automata, 28 of them exhibit random behavior over time, but spatio-temporal currents still lurk underneath.