Minimum description length, regularisation and multi-modal data

Relationships between clustering, description length, and regularisation are pointed out, motivating the introduction of a cost function with a description length interpretation and the unusual and useful property of having its minimum approximated by the densest mode of a distribution. A simple inverse kinematics example is used to demonstrate that this property can be used to select and learn one branch of a multi-valued mapping. This property is also used to develop a method for setting regularisation parameters according to the scale on which structure is exhibited in the training data. The regularisation technique is demonstrated on two real data sets, a classification problem and a regression problem.

Minimum description length, regularisation and multi-modal data

Conventional feed forward Neural Networks have used the sum-of-squares cost function for training. A new cost function is presented here with a description length interpretation based on Rissanen's Minimum Description Length principle. It is a heuristic that has a rough interpretation as the number of data points fit by the model. Not concerned with finding optimal descriptions, the cost function prefers to form minimum descriptions in a naive way for computational convenience. The cost function is called the Naive Description Length cost function. Finding minimum description models will be shown to be closely related to the identification of clusters in the data. As a consequence the minimum of this cost function approximates the most probable mode of the data rather than the sum-of-squares cost function that approximates the mean. The new cost function is shown to provide information about the structure of the data. This is done by inspecting the dependence of the error to the amount of regularisation. This structure provides a method of selecting regularisation parameters as an alternative or supplement to Bayesian methods. The new cost function is tested on a number of multi-valued problems such as a simple inverse kinematics problem. It is also tested on a number of classification and regression problems. The mode-seeking property of this cost function is shown to improve prediction in time series problems. Description length principles are used in a similar fashion to derive a regulariser to control network complexity.

Experimental investigation of inter-modal cross-gain modulation and transient effects in a two mode group erbium doped fiber amplifier

We report what we believe to be the first experimental study of inter-modal cross-gain modulation and associated transient effects as different spatial modes and wavelength channels are added and dropped within a two-mode amplifier for SDM transmission.

Experimental investigation of inter-modal cross-gain modulation and transient effects in a two mode group erbium doped fiber amplifier

We report what we believe to be the first experimental study of inter-modal cross-gain modulation and associated transient effects as different spatial modes and wavelength channels are added and dropped within a two-mode amplifier for SDM transmission.

Imagery in multi-modal object learning

Spatial objects may not only be perceived visually but also by touch. We report recent experiments investigating to what extent prior object knowledge acquired in either the haptic or visual sensory modality transfers to a subsequent visual learning task. Results indicate that even mental object representations learnt in one sensory modality may attain a multi-modal quality. These findings seem incompatible with picture-based reasoning schemas but leave open the possibility of modality-specific reasoning mechanisms.

Gender differences in the sensitivity to negative stimuli:cross-modal affective priming study

Background: There is evidence showing that men and women differ with regard to the processing of emotional information. However, the mechanisms behind these differences are not fully understood. Method: The sample comprised of 275 (167 female) right-handed, healthy participants, recruited from the community. We employed a customized affective priming task, which consisted of three subtests, differing in the modality of the prime (face, written word, and sound). The targets were always written words of either positive or negative valence. The priming effect was measured as reaction time facilitation in conditions where both prime and target were emotional (of the same positive or negative valence) compared with conditions where the emotional targets were preceded by neutral primes. Results: The priming effect was observed across all three modalities, with an interaction of gender by valence: the priming effect in the emotionally negative condition in male participants was stronger compared with females. This was accounted for by the differential priming effect within the female group where priming was significantly smaller in the emotionally negative conditions compared with the positive conditions. The male participants revealed a comparable priming effect across both the emotionally negative and positive conditions. Conclusion: Reduced priming in negative conditions in women may reflect interference processes due to greater sensitivity to negative valence of stimuli. This in turn could underlie the gender-related differences in susceptibility to emotional disorders.

Volume strain sensor based on spectra analysis of in-fiber modal interferometer

An optical in-fiber modal interferometer-based volume strain sensor for earthquake prediction is proposed and experimentally demonstrated. The sensing element is formed by wrapping a multimode-singlemode-multimode fiber structure onto a polyurethane hollow column. Due to the modal interference between the excited guided modes in the fiber, strong interference pattern could be observed in the transmission spectrum. Theoretical analysis verifies that the resonant wavelength shifts as a result of the volume strain variation caused by the column deformation with square root relationship. Sensitivity > 3.93 pm/με within the volume strain ranging from 0 to 1300 με is also experimentally demonstrated. By taking the response of bidirectional change of volume strain and the sluggish character of the employed sensing material into consideration, the sensing system presents good repeatability and stability. © 2001-2012 IEEE.

Impact of inter-modal four-wave mixing on the performance of mode- and wavelength-division-multiplexing systems

In this paper, we investigate the impact of inter-modal four-wave mixing on mode- and wavelength-division-multiplexing systems. A set of coupled nonlinear Schrödinger equations, including linear mode coupling, is derived allowing to isolate the inter-modal four-wave mixing terms. The efficiency of inter-modal four-wave mixing between degenerate LP modes is found to be significantly higher than the intra-modal four-wave mixing efficiency. However, it is shown that the inter-modal four-wave mixing efficiency between degenerate modes is significantly reduced by the linear mode coupling.

Simultaneous measurement for liquid level and refractive index based on all-fiber modal interferometer

A simple fiber sensor capable of simultaneous measurement of liquid level and refractive index (RI) is proposed and experimentally demonstrated. The sensing head is an all-fiber modal interferometer manufactured by splicing an uncoated single-mode fiber with two short sections of multimode fiber. The interference pattern experiences blue shift along with an increase of axial strain and surrounding RI. Owing to the participation of multiple cladding modes with different sensitivities, the height and RI of the liquid could be simultaneously measured by monitoring two dips of the transmission spectrum. Experimental results show that the liquid level and RI sensitivities of the two dips are 245.7 pm/mm, -38 nm/RI unit (RIU), and 223.7 pm/mm, -62 nm/RIU, respectively. The approach has distinctive advantages of easy fabrication, low cost, and high sensitivity for liquid level detection with the capability of distinguishing the RI variation simultaneously. © 2013 Copyright Taylor and Francis Group, LLC.

Pd/C catalysts based on synthetic carbons with bi- and tri-modal pore-size distribution:applications in flow chemistry

Two new types of phenolic resin-derived synthetic carbons with bi-modal and tri-modal pore-size distributions were used as supports for Pd catalysts. The catalysts were tested in chemoselective hydrogenation and hydrodehalogenation reactions in a compact multichannel flow reactor. Bi-modal and tri-modal micro-mesoporous structures of the synthetic carbons were characterised by N2 adsorption. HR-TEM, PXRD and XPS analyses were performed for characterising the synthesised catalysts. N2 adsorption revealed that tri-modal synthetic carbon possesses a well-developed hierarchical mesoporous structure (with 6.5 nm and 42 nm pores), contributing to a larger mesopore volume than the bi-modal carbon (1.57 cm3 g-1versus 1.23 cm3 g-1). It was found that the tri-modal carbon promotes a better size distribution of Pd nanoparticles than the bi-modal carbon due to presence of hierarchical mesopore limitting the growth of Pd nanoparticles. For all the model reactions investigated, the Pd catalyst based on tri-modal synthetic carbon (Pd/triC) show high activity as well as high stability and reproducibility. The trend in reactivities of different functional groups over the Pd/triC catalyst follows a general order alkyne ≫ nitro > bromo ≫ aldehyde.