Raphaël Laurent

Adverse conditions improve distinguishability of auditory, motor and perceptuo-motor theories of speech perception: an exploratory Bayesian modeling study

In this paper, we put forward a computational framework for the comparison between motor, auditory, and perceptuo-motor theories of speech communication. We first recall the basic arguments of these three sets of theories, either applied to speech perception or to speech production. Then we expose a unifying Bayesian model able to express each theory in a probabilistic way. Focusing on speech perception, we demonstrate that under two hypotheses, regarding communication noise and inter-speaker variability, providing perfect conditions for speech communication, motor, and auditory theories are indistinguishable. We then degrade successively each hypothesis to study the distinguishability of the different theories in “adverse” conditions. We first present simulations on a simplified implementation of the model with mono-dimensional sensory and motor variables, and secondly we consider a simulation of the human vocal tract providing more realistic auditory and articulatory variables. Simulation results allow us to emphasise the respective roles of motor and auditory knowledge in various conditions of speech perception in adverse conditions, and to suggest some guidelines for future studies aiming at assessing the role of motor knowledge in speech perception.

Cognitive computation: A Bayesian machine case study

The work presented in this paper is part of the BAMBI project, which aims at better understanding natural cognition by designing non Von Neumann machines with biologicaly plausible hardware. Probabilistic programming allows artificial systems to better operate with uncertainty, and stochastic arithmetic provides a way to carry out approximate computations with few resources. As such, both are plausible models for natural cognition. Our work on the automatic design of probabilistic machines computing soft inferences with an arithmetic based on stochastic bitstreams allowed us to develop the following compilation toolchain: given a high level description of some general problem (typically to infer some knowledge from a model given some observations), formalized as a Bayesian Program, our toolchain automatically builds a low level description of an electronic circuit computing the corresponding probabilistic inference. This circuit can then be implemented and tested on reconfigurable logic.We designed as a validating example a circuit description of a Bayesian filter solving the problem of Pseudo Noise sequence acquisition in telecommunications.

Autonomous robot controller using bitwise gibbs sampling

In the present paper we describe a bio-inspired non von Neumann controller for a simple sensorimotor robotic system. This controller uses a bitwise version of the Gibbs sampling algorithm to select commands so the robot can adapt its course of action and avoid perceived obstacles in the environment. The VHDL specification of the circuit implementation of this controller is based on stochastic computation to perform Bayesian inference at a low energy cost. We show that the proposed unconventional architecture allows to successfully carry out the obstacle avoidance task and to address scalability issues observed in previous works.

Bayesian sensor fusion with fast and low power stochastic circuits

As the physical limits of Moores law are being reached, a research effort is launched to achieve further performance improvements by exploring computation paradigms departing from standard approaches. The BAMBI project (Bottom-up Approaches to Machines dedicated to Bayesian Inference) aims at developing hardware dedicated to probabilistic computation, which extends logic computation realised by boolean gates in current computer chips. Such probabilistic computing devices would allow to solve faster and at a lower energy cost a wide range of Artificial Intelligence applications, especially when decisions need to be taken from incomplete data in an uncertain environment. This paper describes an architecture where very simple operators compute on a time coding of probability values as stochastic signals. Simulation tests and a reconfigurable logic hardware implementation demonstrated the feasibility and performances of the proposed inference machine. Hardware results show this architecture can quickly solve Bayesian sensor fusion problems and is very efficient in terms of energy consumption.

The complementary roles of auditory and motor information evaluated in a Bayesian perceptuo-motor model of speech perception

There is a consensus concerning the view that both auditory and motor representations intervene in the perceptual processing of speech units. However, the question of the functional role of each of these systems remains seldom addressed and poorly understood. We capitalized on the formal framework of Bayesian Programming to develop COSMO (Communicating Objects using Sensory-Motor Operations), an integrative model that allows principled comparisons of purely motor or purely auditory implementations of a speech perception task and tests the gain of efficiency provided by their Bayesian fusion. Here, we show 3 main results: (a) In a set of precisely defined “perfect conditions,” auditory and motor theories of speech perception are indistinguishable; (b) When a learning process that mimics speech development is introduced into COSMO, it departs from these perfect conditions. Then auditory recognition becomes more efficient than motor recognition in dealing with learned stimuli, while motor recognition is more efficient in adverse conditions. We interpret this result as a general “auditory-narrowband versus motor-wideband” property; and (c) Simulations of plosive-vowel syllable recognition reveal possible cues from motor recognition for the invariant specification of the place of plosive articulation in context that are lacking in the auditory pathway. This provides COSMO with a second property, where auditory cues would be more efficient for vowel decoding and motor cues for plosive articulation decoding. These simulations provide several predictions, which are in good agreement with experimental data and suggest that there is natural complementarity between auditory and motor processing within a perceptuo-motor theory of speech perception.

Design of Stochastic Machines Dedicated to Approximate Bayesian inferences

We present an architecture and a compilation toolchain for stochastic machines dedicated to Bayesian inferences. These machines are not Von Neumann and code information with stochastic bitstreams instead of using floating point representations. They only rely on stochastic arithmetic and on Gibbs sampling to perform approximate inferences. They use banks of binary random generators which capture the prior knowledge on which the inference is built. The output of the machine is devised to continuously sample the joint probability distribution of interest. While the method is explained on a simple example, we show that our machine computes a good approximation of the solution to a problem intractable in exact inference.

Cognitive Computation: An Exact Bayesian Inference Stochastic Machine

Probabilisticprogrammingallowsartificialsystemstobetteroperatewithuncertainty,andstochastic arithmeticprovidesawaytocarryoutapproximatecomputationswithfewresources.Assuch,both areplausiblemodelsfornaturalcognition.Theauthors’workontheautomaticdesignofprobabilistic machinescomputingsoftinferences,withanarithmeticbasedonstochasticbitstreams,allowedto developthefollowingcompilationtoolchain:givenahigh-leveldescriptionofsomegeneralproblem, formalized as aBayesianProgram, the toolchain automaticallybuilds a low-level descriptionof anelectroniccircuitcomputingthecorrespondingprobabilisticinference.Thiscircuitcanthenbe implemented and tested on reconfigurable logic. This paper describes two circuits as validating examples.ThefirstoneimplementsaBayesianfiltersolvingtheproblemofPseudoNoisesequence acquisitionintelecommunications.Thesecondoneimplementsdecisionmakinginasensorimotor system:itallowsasimplerobottoavoidobstaclesusingBayesiansensorfusion. KEywoRdS Bayesian Filter, Bayesian Inference, Bayesian Robots, Lfsrs in Telecommunication, Probabilistic Programming, Stochastic Computing

On the Robustness of Stochastic Bayesian Machines

This paper revisits the stochastic computing paradigm as a way to implement architectures dedicated to probabilistic inference. In general, it is assumed the operation over stochastic bit streams is robust with respect to radiation transient events effects. Moreover, it can be expected that leveraging the stochastic computing paradigm to implement probabilistic computations such as Bayesian inference implemented in hardware could yield an increased resilience to radiation effects comparatively to deterministic procedures. However, the practical assessment of the robustness against radiation is mandatory before considering stochastic Bayesian machines (SBMs) in hazardous environments. Results of fault injection campaigns at register transfer level provide the first evidences of the intrinsic robustness of SBMs with respect to single event upsets and single event transients.