Search

Increasing the binding affinity of an antibody to its target antigen is a crucial task in antibody therapeutics development. This paper presents a pretrainable geometric graph neural network, GearBind, and explores its potential in in silico affinity maturation. Leveraging multi-relational graph construction, multi-level geometric message passing and contrastive pretraining on mass-scale, unlabeled protein structural data, GearBind outperforms previous state-of-the-art approaches on SKEMPI and an independent test set. A powerful ensemble model based on GearBind is then derived and used to successfully enhance the binding of two antibodies with distinct formats and target antigens. ELISA EC 50 values of the designed antibody mutants are decreased by up to 17 fold, and K D values by up to 6.1 fold. These promising results underscore the utility of geometric deep learning and effective pretraining in macromolecule interaction modeling tasks., (© 2024. The Author(s).)

When processing language, the brain is thought to deploy specialized computations to construct meaning from complex linguistic structures. Recently, artificial neural networks based on the Transformer architecture have revolutionized the field of natural language processing. Transformers integrate contextual information across words via structured circuit computations. Prior work has focused on the internal representations ("embeddings") generated by these circuits. In this paper, we instead analyze the circuit computations directly: we deconstruct these computations into the functionally-specialized "transformations" that integrate contextual information across words. Using functional MRI data acquired while participants listened to naturalistic stories, we first verify that the transformations account for considerable variance in brain activity across the cortical language network. We then demonstrate that the emergent computations performed by individual, functionally-specialized "attention heads" differentially predict brain activity in specific cortical regions. These heads fall along gradients corresponding to different layers and context lengths in a low-dimensional cortical space., (© 2024. The Author(s).)

An important difference between brains and deep neural networks is the way they learn. Nervous systems learn online where a stream of noisy data points are presented in a non-independent, identically distributed way. Further, synaptic plasticity in the brain depends only on information local to synapses. Deep networks, on the other hand, typically use non-local learning algorithms and are trained in an offline, non-noisy, independent, identically distributed setting. Understanding how neural networks learn under the same constraints as the brain is an open problem for neuroscience and neuromorphic computing. A standard approach to this problem has yet to be established. In this paper, we propose that discrete graphical models that learn via an online maximum a posteriori learning algorithm could provide such an approach. We implement this kind of model in a neural network called the Sparse Quantized Hopfield Network. We show our model outperforms state-of-the-art neural networks on associative memory tasks, outperforms these networks in online, continual settings, learns efficiently with noisy inputs, and is better than baselines on an episodic memory task., (© 2024. The Author(s).)

Inertial Measurement Unit-based methods have great potential in capturing motion in large-scale and complex environments with many people. Sparse Inertial Measurement Unit-based methods have more research value due to their simplicity and flexibility. However, improving the computational efficiency and reducing latency in such methods are challenging. In this paper, we propose Fast Inertial Poser, which is a full body motion estimation deep neural network based on 6 inertial measurement units considering body parameters. We design a network architecture based on recurrent neural networks according to the kinematics tree. This method introduces human body shape information by the causality of observations and eliminates the dependence on future frames. During the estimation of joint positions, the upper body and lower body are estimated using separate network modules independently. Then the joint rotation is obtained through a well-designed single-frame kinematics inverse solver. Experiments show that the method can greatly improve the inference speed and reduce the latency while ensuring the reconstruction accuracy compared with previous methods. Fast Inertial Poser runs at 65 fps with 15 ms latency on an embedded computer, demonstrating the efficiency of the model., (© 2024. The Author(s).)

In order to realize the full potential of wireless edge artificial intelligence (AI), very large and diverse datasets will often be required for energy-demanding model training on resource-constrained edge devices. This paper proposes a lead federated neuromorphic learning (LFNL) technique, which is a decentralized energy-efficient brain-inspired computing method based on spiking neural networks. The proposed technique will enable edge devices to exploit brain-like biophysiological structure to collaboratively train a global model while helping preserve privacy. Experimental results show that, under the situation of uneven dataset distribution among edge devices, LFNL achieves a comparable recognition accuracy to existing edge AI techniques, while substantially reducing data traffic by >3.5× and computational latency by >2.0×. Furthermore, LFNL significantly reduces energy consumption by >4.5× compared to standard federated learning with a slight accuracy loss up to 1.5%. Therefore, the proposed LFNL can facilitate the development of brain-inspired computing and edge AI., (© 2022. The Author(s).)

In this paper we present an adaptive synaptic array that can be used to improve the energy-efficiency of training machine learning (ML) systems. The synaptic array comprises of an ensemble of analog memory elements, each of which is a micro-scale dynamical system in its own right, storing information in its temporal state trajectory. The state trajectories are then modulated by a system level learning algorithm such that the ensemble trajectory is guided towards the optimal solution. We show that the extrinsic energy required for state trajectory modulation can be matched to the dynamics of neural network learning which leads to a significant reduction in energy-dissipated for memory updates during ML training. Thus, the proposed synapse array could have significant implications in addressing the energy-efficiency imbalance between the training and the inference phases observed in artificial intelligence (AI) systems., (© 2022. The Author(s).)

Cluster synchronization in networks of coupled oscillators is the subject of broad interest from the scientific community, with applications ranging from neural to social and animal networks and technological systems. Most of these networks are directed, with flows of information or energy that propagate unidirectionally from given nodes to other nodes. Nevertheless, most of the work on cluster synchronization has focused on undirected networks. Here we characterize cluster synchronization in general directed networks. Our first observation is that, in directed networks, a cluster A of nodes might be one-way dependent on another cluster B: in this case, A may remain synchronized provided that B is stable, but the opposite does not hold. The main contribution of this paper is a method to transform the cluster stability problem in an irreducible form. In this way, we decompose the original problem into subproblems of the lowest dimension, which allows us to immediately detect inter-dependencies among clusters. We apply our analysis to two examples of interest, a human network of violin players executing a musical piece for which directed interactions may be either activated or deactivated by the musicians, and a multilayer neural network with directed layer-to-layer connections.