Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Co-Planar Stereotaxic Atlas of the Human Brain—3-Dimensional Proportional System: An Approach to Cerebral Imaging, J. Talairach, P. Tournoux. Georg Thieme Verlag, New York (1988), 122 pp., 130 figs. DM 268

I have developed "tennis elbow" from lugging this book around the past four weeks, but it is worth the pain, the effort, and the aspirin. It is also worth the (relatively speaking) bargain price. Including appendixes, this book contains 894 pages of text. The entire panorama of the neural sciences is surveyed and examined, and it is comprehensive in its scope, from genomes to social behaviors. The editors explicitly state that the book is designed as "an introductory text for students of biology, behavior, and medicine," but it is hard to imagine any audience, interested in any fragment of neuroscience at any level of sophistication, that would not enjoy this book. The editors have done a masterful job of weaving together the biologic, the behavioral, and the clinical sciences into a single tapestry in which everyone from the molecular biologist to the practicing psychiatrist can find and appreciate his or

Principles of Neural Science

A category of stimuli of great importance for primates, humans in particular, is that formed by actions done by other individuals. If we want to survive, we must understand the actions of others. Furthermore, without action understanding, social organization is impossible. In the case of humans, there is another faculty that depends on the observation of others' actions: imitation learning. Unlike most species, we are able to learn by imitation, and this faculty is at the basis of human culture. In this review we present data on a neurophysiological mechanism--the mirror-neuron mechanism--that appears to play a fundamental role in both action understanding and imitation. We describe first the functional properties of mirror neurons in monkeys. We review next the characteristics of the mirror-neuron system in humans. We stress, in particular, those properties specific to the human mirror-neuron system that might explain the human capacity to learn by imitation. We conclude by discussing the relationship between the mirror-neuron system and language.

The mirror-neuron system.

1. The aim of this study was to find kinematic patterns that are invariant across the normal range of locomotion speeds. Subjects walked at different, freely chosen speeds ranging from 0.9 to 2.1 m s-1, while motion and ground reaction forces on the right side of the body were recorded in three-dimensional space. 2. The time course of the anatomical angles of flexion-extension at the hip and ankle was variable not only across subjects, but even from trial to trial in the same subject. By contrast, the time course of the changes in the angles of elevation of each limb segment (pelvis, thigh, shank and foot) relative to the vertical was stereotyped across subjects. 3. To compare the waveforms across speeds, data were scaled in time relative to gait cycle duration. The pattern of ground reaction forces was highly speed dependent. Several distinct families of curves could be recognized in the flexion-extension angles at the hip and ankle. Instead, the waveforms of global length and elevation of the limb, elevation angles of all limb segments and flexion-extension at the knee were invariant with speed. 4. When gait trajectories at all speeds are plotted in the position space defined by the elevation angles of the limb segments, they describe regular loops on a plane. The statistical characteristics of these angular covariations were quantified by means of principal component analysis. The first two principal components accounted together for > 99% of the total experimental variance, and were quantitatively comparable in all subjects. 5. This constraint of planar covariation of the elevation angles is closely reminiscent of that previously described for the control of posture. The existence of laws of intersegmental co-ordination, common to the control of posture and locomotion, presumably assures the maintenance of dynamic equilibrium during forward progression, and the anticipatory adaptation to potentially destabilizing factors by means of co-ordinated kinematic synergies of the whole body.

Kinematic determinants of human locomotion

http://homes.di.unimi.it/ferrari/publications/journals/2001_Neuroimage.pdf

Different brain correlates for watching real and virtual hand actions.

1. The aim of this study was to describe the time-varying changes in the mechanical parameters of a multijointed limb. The parameters we considered are the coefficients of stiffness, viscosity, and inertia. Continuous pseudorandom perturbations were applied at the elbow joint during a catching task. A modified version of an ensemble technique was used for the identification of time-varying parameters. Torques at the elbow and wrist joints were then modeled with a linear combination of the changes in angular position and velocity weighted by the matrix of angular stiffness and the matrix of angular viscosity, respectively. Control experiments were also performed that involved the stationary maintenance of a given limb posture by resisting actively the applied perturbations. Different limb postures were examined in each such experiment to investigate the dependence of the mechanical parameters on limb geometry. 2. The technique for the identification of limb mechanical parameters proved adequate. The input perturbations applied at the elbow joint elicited angular oscillations at the wrist essentially uncorrelated with those produced at the elbow. The frequency of oscillation is much higher at the wrist than at the elbow, mainly because of the smaller inertia. The variance accounted for by the model was approximately 80% under both stationary and time-varying conditions; in the latter case the value did not vary significantly throughout the task. In addition, the model predicted values of the inertial parameters that were close to the anthropometric measures, and it reproduced the stepwise increase in limb inertia that occurs at the time the ball is held in the hand. 3. The values of angular stiffness and viscosity estimated under stationary conditions did not vary significantly with joint angle, in agreement with previous results obtained under quasi-static postural conditions. The matrix of the coefficients of angular stiffness was not symmetrical, indicating a prominent role for nonautogenic reflex feedbacks with unequal gains for elbow and wrist muscles. 4. A complex temporal modulation of angular stiffness and viscosity was observed during the catching task. The changes in the direct coefficients of angular stiffness tended to covary with those in the coupling coefficients from trial start up to approximately 30 ms before impact time. Around impact time, however, there was a complete dissociation: the direct terms peaked, whereas the coupling terms dropped. The direct terms of angular viscosity also increased before impact, whereas the viscosity coupling terms remained close to zero throughout.(ABSTRACT TRUNCATED AT 400 WORDS)

Time-varying mechanical behavior of multijointed arm in man

1. Load perturbation responses can violate the law of reciprocal innervation between antagonist muscles under particular conditions. Thus flexor and extensor muscles of wrist and elbow joints are reflexly coactivated by the impact of a ball on the hand during a catching task. The aim of this study was to determine whether reflex coactivation can be preset within the central nervous system (CNS) or whether it is entirely due to the peripheral stimulus. To this end, we studied the behavior of stretch reflex responses of arm muscles evoked by torque motor perturbations applied before and during the catching task. 2. Subjects were instructed to catch a ball dropped from 1.6 m. A torque motor delivered perturbations to the elbow joint, resulting in angular motion at both elbow and wrist joints because of their dynamic mechanical coupling. Two series of experiments were performed that differed in the perturbation waveform. In the first series, a single torque pulse could be randomly applied at different times during the task. The corresponding responses were recovered by subtracting the average of the unperturbed trials from the averages of perturbed trials. In the second series of experiments, a train of pseudorandom pulses was applied continuously during each trial. The time-varying impulse responses were computed at 20-ms intervals by cross-correlation methods. 3. The pattern of the short-latency electromyographic responses evoked by either single pulses or pseudorandom perturbations obeyed the law of reciprocal innervation of antagonist muscles under basal conditions. However, the pattern of the responses evoked by the same perturbations around the time of ball impact on the hand consisted of a substantial coactivation of both stretched and shortening muscles. Reflex coactivation resulted from response patterns that differed at different joints. At the elbow, reflex coactivation resulted from a transient reversal of the direction of the short-latency responses of flexor muscles, with little changes of the responses of extensor muscles. At the wrist, instead, reflex coactivation resulted from simultaneous changes in the response waveform of both flexor and extensor muscles. 4. The peripheral conditions associated with the applied perturbations were constant before the time of ball impact. Thus, because the changes of the stretch reflex responses began before that time, they must have been generated within the CNS. It is here hypothesized that the reversal of the reflex responses is centrally gated by switching from the pathways of reciprocal inhibition to those of coactivation of antagonist alpha-motoneurons.(ABSTRACT TRUNCATED AT 400 WORDS)

Transient reversal of the stretch reflex in human arm muscles.

In this paper, we present a novel procedure for the on-the-field autocalibration of triaxial micro accelerometers, which requires neither any equipment nor a controlled environment and allows increasing the accuracy of this kind of microsensor. The procedure exploits the fact that, in static conditions, the modulus of the accelerometer output vector matches that of the gravity acceleration. The calibration model incorporates the bias and scale factor for each axis and the cross-axis symmetrical factors. The parameters are computed through nonlinear optimization, which is solved in a very short time. The calibration procedure was quantitatively tested by comparing the orientation produced by MEMS with that measured by a motion capture system. Results show that the MEMS output, after the calibration procedure, is far more accurate with respect to the output obtained using factory calibration data and almost one order of magnitude more accurate with respect to using traditional calibration models.

Nunzio Alberto Borghese

Papers

Kinematic determinants of human locomotion

Different brain correlates for watching real and virtual hand actions.

Time-varying mechanical behavior of multijointed arm in man

Transient reversal of the stretch reflex in human arm muscles.

Autocalibration of MEMS Accelerometers