Tracking the states of a nonlinear system in the weight-space of a feed-forward neural network

Nonlinear, non-stationary signals are commonly found in a variety of disciplines such as biology, medicine, geology and financial modeling. The complexity (e.g. nonlinearity and non-stationarity) of such signals and their low signal to noise ratios often make it a challenging task to use them in critical applications. In this paper we propose a new neural network based technique to address those problems. We show that a feed forward, multi-layered neural network can conveniently capture the states of a nonlinear system in its connection weight-space, after a process of supervised training. The performance of the proposed method is investigated via computer simulations.

Experimental muscle pain changes feedforward postural responses of the trunk muscles

Many studies have identified changes in trunk muscle recruitment in clinical low back pain (LBP). However, due to the heterogeneity of the LBP population these changes have been variable and it has been impossible to identify a cause-effect relationship. Several studies have identified a consistent change in the feed-forward postural response of transversus abdominis (TrA), the deepest abdominal muscle, in association with arm movements in chronic LBP. This study aimed to determine whether the feedforward recruitment of the trunk muscles in a postural task could be altered by acute experimentally induced LBP. Electromyographic (EMG) recordings of the abdominal and paraspinal muscles were made during arm movements in a control trial, following the injection of isotonic (non-painful) and hypertonic (painful) saline into the longissimus muscle at L4, and during a 1-h follow-up. Movements included rapid arm flexion in response to a light and repetitive arm flexion-extension. Temporal and spatial EMG parameters were measured. The onset and amplitude of EMG of most muscles was changed in a variable manner during the period of experimentally induced pain. However, across movement trials and subjects the activation of TrA was consistently reduced in amplitude or delayed. Analyses in the time and frequency domain were used to confirm these findings. The results suggest that acute experimentally induced pain may affect feedforward postural activity of the trunk muscles. Although the response was variable, pain produced differential changes in the motor control of the trunk muscles, with consistent impairment of TrA activity.

Spatio-temporal evaluation of neck muscle activation during postural perturbations in healthy subjects

The purpose of this study was to examine the spatio-temporal activation of the sternocleidomastoid (SCM) and cervical extensor (CE) muscles with respect to the deltoid muscle onset during rapid voluntary upper limb movement in healthy volunteers. The repeatability and reliability of the spatio-temporal aspects of the myoelectric signals were also examined. Ten subjects performed bilateral and unilateral rapid upper limb flexion, abduction and extension in response to a visual stimulus. EMG onsets and normalised root mean square (nRMS) values were calculated for the SCM and CE muscles. Subjects attended three testing sessions over non-consecutive days allowing the repeatability and reliability of these measures to be assessed. The SCM and CE muscles demonstrated feed-forward activation (activation within 50 ms of deltoid onset) during rapid arm movements in all directions. The sequence and magnitude of neck muscle activation displayed directional specificity, however, the neck flexor and extensor muscles displayed co-activation during all perturbations. EMG onsets demonstrated high repeatability in terms of repeated measure precision (nSEM in the range 1.9-5.7%). This was less evident for the repeatability of nRMS values. The results of this study provide a greater understanding of cervical neuromotor control strategies. During bilateral and unilateral upper limb perturbations, the SCM and CE muscles demonstrate feed-forward co-activation. It seems apparent that feed-forward activation of neck muscles is a mechanism necessary to achieve stability for the visual and vestibular systems, whilst ensuring stabilisation and protection of the cervical spine. (C) 2004 Elsevier Ltd. All rights reserved.

Nonlinear quantum optical computing via measurement

We show how the measurement induced model of quantum computation proposed by Raussendorf and Briegel ( 2001, Phys. Rev. Letts., 86, 5188) can be adapted to a nonlinear optical interaction. This optical implementation requires a Kerr nonlinearity, a single photon source, a single photon detector and fast feed forward. Although nondeterministic optical quantum information proposals such as that suggested by KLM ( 2001, Nature, 409, 46) do not require a Kerr nonlinearity they do require complex reconfigurable optical networks. The proposal in this paper has the benefit of a single static optical layout with fixed device parameters, where the algorithm is defined by the final measurement procedure.

Delayed onset of transversus abdominus in long-standing groin pain

Long-standing groin pain is a persistent problem that is commonly difficult to rehabilitate. Theoretical rationale indicates a relationship between the motor control of the pelvis and long-standing groin pain; however, this link has not been investigated. Purpose: The current experiment aimed to evaluate motor control of the abdominal muscles in a group of Australian football players with and without long-standing groin pain. Methods: Ten participants with long-standing groin pain and 12 asymptomatic controls were recruited for the study. Participants were elite or subelite Australian football players. Fine-wire and surface electromyography electrodes were used to record the activity of the selected abdominal and leg muscles during a visual choice reaction-time task (active straight leg raising). Results: When the asymptomatic controls completed the active straight leg raise (ASLR) task, the transversus abdominus contracted in a feed-forward manner. However, when individuals with long-standing groin pain completed the ASLR task, the onset of transversus abdominus was delayed (P < 0.05) compared with the control group. There were no differences between groups for the onset of activity of internal oblique, external oblique, and rectus abdominus (all P > 0.05). Conclusions: The finding that the onset of transversus abdominus is delayed in individuals with long-standing groin pain is important, as it demonstrates an association between long-standing groin pain and transversus abdominus activation.

Optimal photons for quantum-information processing

Photonic quantum-information processing schemes, such as linear optics quantum computing, and other experiments relying on single-photon interference, inherently require complete photon indistinguishability to enable the desired photonic interactions to take place. Mode-mismatch is the dominant cause of photon distinguishability in optical circuits. Here we study the effects of photon wave-packet shape on tolerance against the effects of mode mismatch in linear optical circuits, and show that Gaussian distributed photons with large bandwidth are optimal. The result is general and holds for arbitrary linear optical circuits, including ones which allow for postselection and classical feed forward. Our findings indicate that some single photon sources, frequently cited for their potential application to quantum-information processing, may in fact be suboptimal for such applications.

Prediction of subcellular localization using sequence-biased recurrent networks

Motivation: Targeting peptides direct nascent proteins to their specific subcellular compartment. Knowledge of targeting signals enables informed drug design and reliable annotation of gene products. However, due to the low similarity of such sequences and the dynamical nature of the sorting process, the computational prediction of subcellular localization of proteins is challenging. Results: We contrast the use of feed forward models as employed by the popular TargetP/SignalP predictors with a sequence-biased recurrent network model. The models are evaluated in terms of performance at the residue level and at the sequence level, and demonstrate that recurrent networks improve the overall prediction performance. Compared to the original results reported for TargetP, an ensemble of the tested models increases the accuracy by 6 and 5% on non-plant and plant data, respectively.

The applicability of recurrent neural networks for biological sequence analysis

Selection of machine learning techniques requires a certain sensitivity to the requirements of the problem. In particular, the problem can be made more tractable by deliberately using algorithms that are biased toward solutions of the requisite kind. In this paper, we argue that recurrent neural networks have a natural bias toward a problem domain of which biological sequence analysis tasks are a subset. We use experiments with synthetic data to illustrate this bias. We then demonstrate that this bias can be exploitable using a data set of protein sequences containing several classes of subcellular localization targeting peptides. The results show that, compared with feed forward, recurrent neural networks will generally perform better on sequence analysis tasks. Furthermore, as the patterns within the sequence become more ambiguous, the choice of specific recurrent architecture becomes more critical.