Modeling Efficient Classification as a Process of Confidence Assessment and Delegation

In visual object detection and recognition, classifiers have two interesting characteristics: accuracy and speed. Accuracy depends on the complexity of the image features and classifier decision surfaces. Speed depends on the hardware and the computational effort required to use the features and decision surfaces. When attempts to increase accuracy lead to increases in complexity and effort, it is necessary to ask how much are we willing to pay for increased accuracy. For example, if increased computational effort implies quickly diminishing returns in accuracy, then those designing inexpensive surveillance applications cannot aim for maximum accuracy at any cost. It becomes necessary to find trade-offs between accuracy and effort. We study efficient classification of images depicting real-world objects and scenes. Classification is efficient when a classifier can be controlled so that the desired trade-off between accuracy and effort (speed) is achieved and unnecessary computations are avoided on a per input basis. A framework is proposed for understanding and modeling efficient classification of images. Classification is modeled as a tree-like process. In designing the framework, it is important to recognize what is essential and to avoid structures that are narrow in applicability. Earlier frameworks are lacking in this regard. The overall contribution is two-fold. First, the framework is presented, subjected to experiments, and shown to be satisfactory. Second, certain unconventional approaches are experimented with. This allows the separation of the essential from the conventional. To determine if the framework is satisfactory, three categories of questions are identified: trade-off optimization, classifier tree organization, and rules for delegation and confidence modeling. Questions and problems related to each category are addressed and empirical results are presented. For example, related to trade-off optimization, we address the problem of computational bottlenecks that limit the range of trade-offs. We also ask if accuracy versus effort trade-offs can be controlled after training. For another example, regarding classifier tree organization, we first consider the task of organizing a tree in a problem-specific manner. We then ask if problem-specific organization is necessary.

Learning Nonlinear Visual Processing from Natural Images

The paradigm of computational vision hypothesizes that any visual function -- such as the recognition of your grandparent -- can be replicated by computational processing of the visual input. What are these computations that the brain performs? What should or could they be? Working on the latter question, this dissertation takes the statistical approach, where the suitable computations are attempted to be learned from the natural visual data itself. In particular, we empirically study the computational processing that emerges from the statistical properties of the visual world and the constraints and objectives specified for the learning process. This thesis consists of an introduction and 7 peer-reviewed publications, where the purpose of the introduction is to illustrate the area of study to a reader who is not familiar with computational vision research. In the scope of the introduction, we will briefly overview the primary challenges to visual processing, as well as recall some of the current opinions on visual processing in the early visual systems of animals. Next, we describe the methodology we have used in our research, and discuss the presented results. We have included some additional remarks, speculations and conclusions to this discussion that were not featured in the original publications. We present the following results in the publications of this thesis. First, we empirically demonstrate that luminance and contrast are strongly dependent in natural images, contradicting previous theories suggesting that luminance and contrast were processed separately in natural systems due to their independence in the visual data. Second, we show that simple cell -like receptive fields of the primary visual cortex can be learned in the nonlinear contrast domain by maximization of independence. Further, we provide first-time reports of the emergence of conjunctive (corner-detecting) and subtractive (opponent orientation) processing due to nonlinear projection pursuit with simple objective functions related to sparseness and response energy optimization. Then, we show that attempting to extract independent components of nonlinear histogram statistics of a biologically plausible representation leads to projection directions that appear to differentiate between visual contexts. Such processing might be applicable for priming, \ie the selection and tuning of later visual processing. We continue by showing that a different kind of thresholded low-frequency priming can be learned and used to make object detection faster with little loss in accuracy. Finally, we show that in a computational object detection setting, nonlinearly gain-controlled visual features of medium complexity can be acquired sequentially as images are encountered and discarded. We present two online algorithms to perform this feature selection, and propose the idea that for artificial systems, some processing mechanisms could be selectable from the environment without optimizing the mechanisms themselves. In summary, this thesis explores learning visual processing on several levels. The learning can be understood as interplay of input data, model structures, learning objectives, and estimation algorithms. The presented work adds to the growing body of evidence showing that statistical methods can be used to acquire intuitively meaningful visual processing mechanisms. The work also presents some predictions and ideas regarding biological visual processing.

Neuronal oscillations in gamma- and alpha-frequency bands: from object representations to sensory awareness

The synchronization of neuronal activity, especially in the beta- (14-30 Hz) /gamma- (30 80 Hz) frequency bands, is thought to provide a means for the integration of anatomically distributed processing and for the formation of transient neuronal assemblies. Thus non-stimulus locked (i.e. induced) gamma-band oscillations are believed to underlie feature binding and the formation of neuronal object representations. On the other hand, the functional roles of neuronal oscillations in slower theta- (4 8 Hz) and alpha- (8 14 Hz) frequency bands remain controversial. In addition, early stimulus-locked activity has been largely ignored, as it is believed to reflect merely the physical properties of sensory stimuli. With human neuromagnetic recordings, both the functional roles of gamma- and alpha-band oscillations and the significance of early stimulus-locked activity in neuronal processing were examined in this thesis. Study I of this thesis shows that even the stimulus-locked (evoked) gamma oscillations were sensitive to high-level stimulus features for speech and non-speech sounds, suggesting that they may underlie the formation of early neuronal object representations for stimuli with a behavioural relevance. Study II shows that neuronal processing for consciously perceived and unperceived stimuli differed as early as 30 ms after stimulus onset. This study also showed that the alpha band oscillations selectively correlated with conscious perception. Study III, in turn, shows that prestimulus alpha-band oscillations influence the subsequent detection and processing of sensory stimuli. Further, in Study IV, we asked whether phase synchronization between distinct frequency bands is present in cortical circuits. This study revealed prominent task-sensitive phase synchrony between alpha and beta/gamma oscillations. Finally, the implications of Studies II, III, and IV to the broader scientific context are analysed in the last study of this thesis (V). I suggest, in this thesis that neuronal processing may be extremely fast and that the evoked response is important for cognitive processes. I also propose that alpha oscillations define the global neuronal workspace of perception, action, and consciousness and, further, that cross-frequency synchronization is required for the integration of neuronal object representations into global neuronal workspace.

Thin-Layer Chromatography with Ultraviolet and Mass Spectrometric Detection: From Preparative-Layer to Miniaturized Ultra-Thin-Layer Technique

The present challenge in drug discovery is to synthesize new compounds efficiently in minimal time. The trend is towards carefully designed and well-characterized compound libraries because fast and effective synthesis methods easily produce thousands of new compounds. The need for rapid and reliable analysis methods is increased at the same time. Quality assessment, including the identification and purity tests, is highly important since false (negative or positive) results, for instance in tests of biological activity or determination of early-ADME parameters in vitro (the pharmacokinetic study of drug absorption, distribution, metabolism, and excretion), must be avoided. This thesis summarizes the principles of classical planar chromatographic separation combined with ultraviolet (UV) and mass spectrometric (MS) detection, and introduces powerful, rapid, easy, low-cost, and alternative tools and techniques for qualitative and quantitative analysis of small drug or drug-like molecules. High performance thin-layer chromatography (HPTLC) was introduced and evaluated for fast semi-quantitative assessment of the purity of synthesis target compounds. HPTLC methods were compared with the liquid chromatography (LC) methods. Electrospray ionization mass spectrometry (ESI MS) and atmospheric pressure matrix-assisted laser desorption/ionization MS (AP MALDI MS) were used to identify and confirm the product zones on the plate. AP MALDI MS was rapid, and easy to carry out directly on the plate without scraping. The PLC method was used to isolate target compounds from crude synthesized products and purify them for bioactivity and preliminary ADME tests. Ultra-thin-layer chromatography (UTLC) with AP MALDI MS and desorption electrospray ionization mass spectrometry (DESI MS) was introduced and studied for the first time. Because of the thinner adsorbent layer, the monolithic UTLC plate provided 10 100 times better sensitivity in MALDI analysis than did HPTLC plates. The limits of detection (LODs) down to low picomole range were demonstrated for UTLC AP MALDI and UTLC DESI MS. In a comparison of AP and vacuum MALDI MS detection for UTLC plates, desorption from the irregular surface of the plates with the combination of an external AP MALDI ion source and an ion trap instrument provided clearly less variation in mass accuracy than the vacuum MALDI time-of-flight (TOF) instrument. The performance of the two-dimensional (2D) UTLC separation with AP MALDI MS method was studied for the first time. The influence of the urine matrix on the separation and the repeatability was evaluated with benzodiazepines as model substances in human urine. The applicability of 2D UTLC AP MALDI MS was demonstrated in the detection of metabolites in an authentic urine sample.

Sensory auditory processing and intuitive sound detection : an investigation of musical experts and nonexperts

The auditory system can detect occasional changes (deviants) in acoustic regularities without the need for subjects to focus their attention on the sound material. Deviant detection is reflected in the elicitation of the mismatch negativity component (MMN) of the event-related potentials. In the studies presented in this thesis, the MMN is used to investigate the auditory abilities for detecting similarities and regularities in sound streams. To investigate the limits of these processes, professional musicians have been tested in some of the studies. The results show that auditory grouping is already more advanced in musicians than in nonmusicians and that the auditory system of musicians can, unlike that of nonmusicians, detect a numerical regularity of always four tones in a series. These results suggest that sensory auditory processing in musicians is not only a fine tuning of universal abilities, but is also qualitatively more advanced than in nonmusicians. In addition, the relationship between the auditory change-detection function and perception is examined. It is shown that, contrary to the generally accepted view, MMN elicitation does not necessarily correlate with perception. The outcome of the auditory change-detection function can be implicit and the implicit knowledge of the sound structure can, after training, be utilized for behaviorally correct intuitive sound detection. These results illustrate the automatic character of the sensory change detection function.

Dynamics of contour, object and face processing in the human visual cortex

The neural basis of visual perception can be understood only when the sequence of cortical activity underlying successful recognition is known. The early steps in this processing chain, from retina to the primary visual cortex, are highly local, and the perception of more complex shapes requires integration of the local information. In Study I of this thesis, the progression from local to global visual analysis was assessed by recording cortical magnetoencephalographic (MEG) responses to arrays of elements that either did or did not form global contours. The results demonstrated two spatially and temporally distinct stages of processing: The first, emerging 70 ms after stimulus onset around the calcarine sulcus, was sensitive to local features only, whereas the second, starting at 130 ms across the occipital and posterior parietal cortices, reflected the global configuration. To explore the links between cortical activity and visual recognition, Studies II III presented subjects with recognition tasks of varying levels of difficulty. The occipito-temporal responses from 150 ms onwards were closely linked to recognition performance, in contrast to the 100-ms mid-occipital responses. The averaged responses increased gradually as a function of recognition performance, and further analysis (Study III) showed the single response strengths to be graded as well. Study IV addressed the attention dependence of the different processing stages: Occipito-temporal responses peaking around 150 ms depended on the content of the visual field (faces vs. houses), whereas the later and more sustained activity was strongly modulated by the observers attention. Hemodynamic responses paralleled the pattern of the more sustained electrophysiological responses. Study V assessed the temporal processing capacity of the human object recognition system. Above sufficient luminance, contrast and size of the object, the processing speed was not limited by such low-level factors. Taken together, these studies demonstrate several distinct stages in the cortical activation sequence underlying the object recognition chain, reflecting the level of feature integration, difficulty of recognition, and direction of attention.

Feature integration in human vision

The earliest stages of human cortical visual processing can be conceived as extraction of local stimulus features. However, more complex visual functions, such as object recognition, require integration of multiple features. Recently, neural processes underlying feature integration in the visual system have been under intensive study. A specialized mid-level stage preceding the object recognition stage has been proposed to account for the processing of contours, surfaces and shapes as well as configuration. This thesis consists of four experimental, psychophysical studies on human visual feature integration. In two studies, classification image a recently developed psychophysical reverse correlation method was used. In this method visual noise is added to near-threshold stimuli. By investigating the relationship between random features in the noise and observer s perceptual decision in each trial, it is possible to estimate what features of the stimuli are critical for the task. The method allows visualizing the critical features that are used in a psychophysical task directly as a spatial correlation map, yielding an effective "behavioral receptive field". Visual context is known to modulate the perception of stimulus features. Some of these interactions are quite complex, and it is not known whether they reflect early or late stages of perceptual processing. The first study investigated the mechanisms of collinear facilitation, where nearby collinear Gabor flankers increase the detectability of a central Gabor. The behavioral receptive field of the mechanism mediating the detection of the central Gabor stimulus was measured by the classification image method. The results show that collinear flankers increase the extent of the behavioral receptive field for the central Gabor, in the direction of the flankers. The increased sensitivity at the ends of the receptive field suggests a low-level explanation for the facilitation. The second study investigated how visual features are integrated into percepts of surface brightness. A novel variant of the classification image method with brightness matching task was used. Many theories assume that perceived brightness is based on the analysis of luminance border features. Here, for the first time this assumption was directly tested. The classification images show that the perceived brightness of both an illusory Craik-O Brien-Cornsweet stimulus and a real uniform step stimulus depends solely on the border. Moreover, the spatial tuning of the features remains almost constant when the stimulus size is changed, suggesting that brightness perception is based on the output of a single spatial frequency channel. The third and fourth studies investigated global form integration in random-dot Glass patterns. In these patterns, a global form can be immediately perceived, if even a small proportion of random dots are paired to dipoles according to a geometrical rule. In the third study the discrimination of orientation structure in highly coherent concentric and Cartesian (straight) Glass patterns was measured. The results showed that the global form was more efficiently discriminated in concentric patterns. The fourth study investigated how form detectability depends on the global regularity of the Glass pattern. The local structure was either Cartesian or curved. It was shown that randomizing the local orientation deteriorated the performance only with the curved pattern. The results give support for the idea that curved and Cartesian patterns are processed in at least partially separate neural systems.

Concepts and Object-Oriented Knowledge Representation

Tutkielma käsittelee nykyisiä kognitiotieteen teorioita käsitteistä ja niiden mallintamista oliokeskeisillä tietämyksen esittämisen menetelmillä. Käsiteteorioista käsitellään klassinen, määritelmäteoria, prototyyppiteoria, duaaliteoriat, uusklassinen teoria, teoria-teoria ja atomistinen teoria. Oliokeskeiset menetelmät ovat viime aikoina jakautuneet kahden tyyppisiin kieliin: oliopohjaisiin ja luokkapohjaisiin. Uudet olio-pohjaiset olio-ohjelmointikielet antavat käsitteiden representointiin mahdollisuuksia, jotka puuttuvat aikaisemmista luokka-pohjaisista kielistä ja myös kehysmenetelmistä. Tutkielma osoittaa, että oliopohjaisten kielten uudet piirteet tarjoavat keinoja, joilla käsitteitä voidaan esittää symbolisessa muodossa paremmin kuin perinteisillä menetelmillä. Niillä pystytään simuloimaan kaikkea mitä luokkapohjaisilla kielillä voidaan, mutta ne pystyvät lisäksi simuloimaan perheyhtäläisyyskäsitteitä ja mahdollistavat olioiden dynaamisen muuttamisen ilman, että siinä rikotaan psykologisen essentialismin periaatetta. Tutkielma osoittaa lisäksi vakavia puutteitta, jotka koskevat koko oliokeskeistä menetelmää. Avainsanat: käsitteet, käsiteteoriat, tekoäly, komputationaalinen psykologia, olio-ohjelmointi, tiedon esittäminen

Early detection of type 2 diabetes mellitus in Chinese and Indian adult population

Objectives of this study were to determine secular trends of diabetes prevalence in China and develop simple risk assessment algorithms for screening individuals with high-risk for diabetes or with undiagnosed diabetes in Chinese and Indian adults. Two consecutive population based surveys in Chinese and a prospective study in Mauritian Indians were involved in this study. The Chinese surveys were conducted in randomly selected populations aged 20-74 years in 2001-2002 (n=14 592) and 35-74 years in 2006 (n=4416). A two-step screening strategy using fasting capillary plasma glucose (FCG) as first-line screening test followed by standard 2-hour 75g oral glucose tolerance tests (OGTTs) was applied to 12 436 individuals in 2001, while OGTTs were administrated to all participants together with FCG in 2006 and to 2156 subjects in 2002. In Mauritius, two consecutive population based surveys were conducted in Mauritian Indians aged 20-65 years in 1987 and 1992; 3094 Indians (1141 men), who were not diagnosed as diabetes at baseline, were reexamined with OGTTs in 1992 and/or 1998. Diabetes and pre-diabetes was defined following 2006 World Health Organization/ International Diabetes Federation Criteria. Age-standardized, as well as age- and sex-specific, prevalence of diabetes and pre-diabetes in adult Chinese was significantly increased from 12.2% and 15.4% in 2001 to 16.0% and 21.2% in 2006, respectively. A simple Chinese diabetes risk score was developed based on the data of Chinese survey 2001-2002 and validated in the population of survey 2006. The risk scores based on β coefficients derived from the final Logistic regression model ranged from 3 – 32. When the score was applied to the population of survey 2006, the area under operating characteristic curve (AUC) of the score for screening undiagnosed diabetes was 0.67 (95% CI, 0.65-0.70), which was lower than the AUC of FCG (0.76 [0.74-0.79]), but similar to that of HbA1c (0.68 [0.65-0.71]). At a cut-off point of 14, the sensitivity and specificity of the risk score in screening undiagnosed diabetes was 0.84 (0.81-0.88) and 0.40 (0.38-0.41). In Mauritian Indian, body mass index (BMI), waist girth, family history of diabetes (FH), and glucose was confirmed to be independent risk predictors for developing diabetes. Predicted probabilities for developing diabetes derived from a simple Cox regression model fitted with sex, FH, BMI and waist girth ranged from 0.05 to 0.64 in men and 0.03 to 0.49 in women. To predict the onset of diabetes, the AUC of the predicted probabilities was 0.62 (95% CI, 0.56-0.68) in men and 0.64(0.59-0.69) in women. At a cut-off point of 0.12, the sensitivity and specificity was 0.72(0.71-0.74) and 0.47(0.45-0.49) in men; and 0.77(0.75-0.78) and 0.50(0.48-0.52) in women, respectively. In conclusion, there was a rapid increase in prevalence of diabetes in Chinese adults from 2001 to 2006. The simple risk assessment algorithms based on age, obesity and family history of diabetes showed a moderate discrimination of diabetes from non-diabetes, which may be used as first line screening tool for diabetes and pre-diabetes, and for health promotion purpose in Chinese and Indians.

Automatic Lameness Detection in a Milking Robot : Instrumentation, measurement software, algorithms for data analysis and a neural network model

The aim of this thesis is to develop a fully automatic lameness detection system that operates in a milking robot. The instrumentation, measurement software, algorithms for data analysis and a neural network model for lameness detection were developed. Automatic milking has become a common practice in dairy husbandry, and in the year 2006 about 4000 farms worldwide used over 6000 milking robots. There is a worldwide movement with the objective of fully automating every process from feeding to milking. Increase in automation is a consequence of increasing farm sizes, the demand for more efficient production and the growth of labour costs. As the level of automation increases, the time that the cattle keeper uses for monitoring animals often decreases. This has created a need for systems for automatically monitoring the health of farm animals. The popularity of milking robots also offers a new and unique possibility to monitor animals in a single confined space up to four times daily. Lameness is a crucial welfare issue in the modern dairy industry. Limb disorders cause serious welfare, health and economic problems especially in loose housing of cattle. Lameness causes losses in milk production and leads to early culling of animals. These costs could be reduced with early identification and treatment. At present, only a few methods for automatically detecting lameness have been developed, and the most common methods used for lameness detection and assessment are various visual locomotion scoring systems. The problem with locomotion scoring is that it needs experience to be conducted properly, it is labour intensive as an on-farm method and the results are subjective. A four balance system for measuring the leg load distribution of dairy cows during milking in order to detect lameness was developed and set up in the University of Helsinki Research farm Suitia. The leg weights of 73 cows were successfully recorded during almost 10,000 robotic milkings over a period of 5 months. The cows were locomotion scored weekly, and the lame cows were inspected clinically for hoof lesions. Unsuccessful measurements, caused by cows standing outside the balances, were removed from the data with a special algorithm, and the mean leg loads and the number of kicks during milking was calculated. In order to develop an expert system to automatically detect lameness cases, a model was needed. A probabilistic neural network (PNN) classifier model was chosen for the task. The data was divided in two parts and 5,074 measurements from 37 cows were used to train the model. The operation of the model was evaluated for its ability to detect lameness in the validating dataset, which had 4,868 measurements from 36 cows. The model was able to classify 96% of the measurements correctly as sound or lame cows, and 100% of the lameness cases in the validation data were identified. The number of measurements causing false alarms was 1.1%. The developed model has the potential to be used for on-farm decision support and can be used in a real-time lameness monitoring system.

Quantitative trait loci detection and benefits from marker-assisted selection in dairy cattle

Knowing the chromosomal areas or actual genes affecting the traits under selection would add more information to be used in the selection decisions which would potentially lead to higher genetic response. The first objective of this study was to map quantitative trait loci (QTL) affecting economically important traits in the Finnish Ayrshire population. The second objective was to investigate the effects of using QTL information in marker-assisted selection (MAS) on the genetic response and the linkage disequilibrium between the different parts of the genome. Whole genome scans were carried out on a grand-daughter design with 12 half-sib families and a total of 493 sons. Twelve different traits were studied: milk yield, protein yield, protein content, fat yield, fat content, somatic cell score (SCS), mastitis treatments, other veterinary treatments, days open, fertility treatments, non-return rate, and calf mortality. The average spacing of the typed markers was 20 cM with 2 to 14 markers per chromosome. Associations between markers and traits were analyzed with multiple marker regression. Significance was determined by permutation and genome-wise P-values obtained by Bonferroni correction. The benefits from MAS were investigated by simulation: a conventional progeny testing scheme was compared to a scheme where QTL information was used within families to select among full-sibs in the male path. Two QTL on different chromosomes were modelled. The effects of different starting frequencies of the favourable alleles and different size of the QTL effects were evaluated. A large number of QTL, 48 in total, were detected at 5% or higher chromosome-wise significance. QTL for milk production were found on 8 chromosomes, for SCS on 6, for mastitis treatments on 1, for other veterinary treatments on 5, for days open on 7, for fertility treatments on 7, for calf mortality on 6, and for non-return rate on 2 chromosomes. In the simulation study the total genetic response was faster with MAS than with conventional selection and the advantage of MAS persisted over the studied generations. The rate of response and the difference between the selection schemes reflected clearly the changes in allele frequencies of the favourable QTL. The disequilibrium between the polygenes and QTL was always negative and it was larger with larger QTL size. The disequilibrium between the two QTL was larger with QTL of large effect and it was somewhat larger with MAS for scenarios with starting frequencies below 0.5 for QTL of moderate size and below 0.3 for large QTL. In conclusion, several QTL affecting economically important traits of dairy cattle were detected. Further studies are needed to verify these QTL, check their presence in the present breeding population, look for pleiotropy and fine map the most interesting QTL regions. The results of the simulation studies show that using MAS together with embryo transfer to pre-select young bulls within families is a useful approach to increase the genetic merit of the AI-bulls compared to conventional selection.