A hardware-software codesign framework for cellular computing

Until recently, the ever-increasing demand of computing power has been met on one hand by increasing the operating frequency of processors and on the other hand by designing architectures capable of exploiting parallelism at the instruction level through hardware mechanisms such as super-scalar execution. However, both these approaches seem to have reached a plateau, mainly due to issues related to design complexity and cost-effectiveness. To face the stabilization of performance of single-threaded processors, the current trend in processor design seems to favor a switch to coarser-grain parallelization, typically at the thread level. In other words, high computational power is achieved not only by a single, very fast and very complex processor, but through the parallel operation of several processors, each executing a different thread. Extrapolating this trend to take into account the vast amount of on-chip hardware resources that will be available in the next few decades (either through further shrinkage of silicon fabrication processes or by the introduction of molecular-scale devices), together with the predicted features of such devices (e.g., the impossibility of global synchronization or higher failure rates), it seems reasonable to foretell that current design techniques will not be able to cope with the requirements of next-generation electronic devices and that novel design tools and programming methods will have to be devised. A tempting source of inspiration to solve the problems implied by a massively parallel organization and inherently error-prone substrates is biology. In fact, living beings possess characteristics, such as robustness to damage and self-organization, which were shown in previous research as interesting to be implemented in hardware. For instance, it was possible to realize relatively simple systems, such as a self-repairing watch. Overall, these bio-inspired approaches seem very promising but their interest for a wider audience is problematic because their heavily hardware-oriented designs lack some of the flexibility achievable with a general purpose processor. In the context of this thesis, we will introduce a processor-grade processing element at the heart of a bio-inspired hardware system. This processor, based on a single-instruction, features some key properties that allow it to maintain the versatility required by the implementation of bio-inspired mechanisms and to realize general computation. We will also demonstrate that the flexibility of such a processor enables it to be evolved so it can be tailored to different types of applications. In the second half of this thesis, we will analyze how the implementation of a large number of these processors can be used on a hardware platform to explore various bio-inspired mechanisms. Based on an extensible platform of many FPGAs, configured as a networked structure of processors, the hardware part of this computing framework is backed by an open library of software components that provides primitives for efficient inter-processor communication and distributed computation. We will show that this dual software–hardware approach allows a very quick exploration of different ways to solve computational problems using bio-inspired techniques. In addition, we also show that the flexibility of our approach allows it to exploit replication as a solution to issues that concern standard embedded applications.

A hardware-software design framework for distributed cellular computing

Michel Ganguin, Pierre-André Mudry, Julien Ruffin

In this article, we describe a novel hardware-software design framework for prototyping cellular architectures in hardware. Based on an extensible platform of about 200 FPGAs, configured as a networked structure of processors, the hardware part of this computing framework is backed by an extensible library of software components that provides primitives for efficient inter-processor communication and distributed computation. This dual software-hardware approach allows a very quick exploration of different ways to solve computational problems using bio-inspired techniques. To demonstrate the validity of the method, we present an example of how a traditional parallel system such as a cellular automaton can be modeled and run with this perspective. In addition, we also show that the flexibility of our approach allows not only cellular automata but any computation to be easily implemented on a cellular substrate.

2008

Towards a theoretical foundation for morphological computation with compliant bodies

Auke Ijspeert, Rolf Pfeifer

The control of compliant robots is, due to their often nonlinear and complex dynamics, inherently difficult. The vision of morphological computation proposes to view these aspects not only as problems, but rather also as parts of the solution. Non-rigid body parts are not seen anymore as imperfect realizations of rigid body parts, but rather as potential computational resources. The applicability of this vision has already been demonstrated for a variety of complex robot control problems. Nevertheless, a theoretical basis for understanding the capabilities and limitations of morphological computation has been missing so far. We present a model for morphological computation with compliant bodies, where a precise mathematical characterization of the potential computational contribution of a complex physical body is feasible. The theory suggests that complexity and nonlinearity, typically unwanted properties of robots, are desired features in order to provide computational power. We demonstrate that simple generic models of physical bodies, based on mass-spring systems, can be used to implement complex nonlinear operators. By adding a simple readout (which is static and linear) to the morphology, such devices are able to emulate complex mappings of input to output streams in continuous time. Hence, by outsourcing parts of the computation to the physical body, the difficult problem of learning to control a complex body, could be reduced to a simple and perspicuous learning task, which can not get stuck in local minima of an error function.

2011

Towards goal-driven deep neural network models to elucidate human arm proprioception

Pranav Mamidanna, Alexander Mathis

Proprioceptive signals are a critical component of our ability to perform complex movements, identify our posture and adapt to environmental changes. Our movements are generated by a large number of muscles and are sensed via a myriad of different receptor types. Even the most important ones, muscle spindles, carry highly multiplexed information. For instance, arm movements are sensed via distributed and individually ambiguous activity patterns of muscle spindles, which depend on relative joint configurations rather than the absolute hand position. This high dimensional input (~50 muscles for a human arm) of distributed information poses a challenging decoding problem for the nervous system. Given the diversity in muscle activity, what are the necessary computations that the proprioceptive system needs to perform to sense our movements? Here we studied a proprioceptive variant of the handwritten character recognition task to gain insight into potential computations that the proprioceptive system needs to perform. We focussed on handwritten character classification of muscle length configuration patterns that were required to draw that character. We started from a dataset comprising of pen-tip trajectory data recorded while subjects were writing individual single-stroke characters of the Latin alphabet (Williams et al. ICANN 2006), and employed a musculoskeletal model of the human upper limb to generate muscle length configurations corresponding to drawing the pen-tip trajectories in multiple horizontal and vertical planes. Using this model we created a large, scalable dataset of muscle length configurations corresponding to handwritten characters of varying sizes, locations and orientations (n > 105 samples). To determine the difficulty of this problem, we trained support vector machines (SVM) to solve a binary one-vs-all classification task on the dataset, which achieves an accuracy of 0.89 ± 0.08 (mean ± s.e.m). Contrary to naive expectation, reading out the character at the level of muscles is much easier whereas SVMs trained on the same task using pen-tip coordinates performed relatively poorly: 0.75 ± 0.14. This suggests that the musculoskeletal system itself serves as a non-linear projection to a higher dimensional space, which simplifies character recognition. Next we focussed on goal-driven deep neural network architectures to achieve higher accuracy. Training deep neural networks requires a large, diverse datasets and challenging tasks. We found that the scalable dataset for character recognition we generated is large enough to constrain deep convolutional architectures. We are currently exploring the performance of different deeplearning architectures in solving the handwritten character classification task to investigate which representations are learned and what computations are most efficient. We found that convolutional neural networks factoring out temporal and inter-muscle ('spatial') information achieve almost perfect accuracy for the multi-class problem. These preliminary results suggest that neural networks can learn pose-invariant character recognition from muscle configurations.

2018

Towards goal-driven deep neural network models to elucidate human arm proprioception

Pranav Mamidanna, Alexander Mathis

2018