Browsing by Author "Kaski, Samuel"

Filter results by typing the first few letters
Now showing 1 - 20 of 208
  • ABC of the future
    (2023-08) Pesonen, Henri; Simola, Umberto; Köhn-Luque, Alvaro; Vuollekoski, Henri; Lai, Xiaoran; Frigessi, Arnoldo; Kaski, Samuel; Frazier, David T.; Maneesoonthorn, Worapree; Martin, Gael M.; Corander, Jukka
     A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä
    Approximate Bayesian computation (ABC) has advanced in two decades from a seminal idea to a practically applicable inference tool for simulator-based statistical models, which are becoming increasingly popular in many research domains. The computational feasibility of ABC for practical applications has been recently boosted by adopting techniques from machine learning to build surrogate models for the approximate likelihood or posterior and by the introduction of a general-purpose software platform with several advanced features, including automated parallelisation. Here we demonstrate the strengths of the advances in ABC by going beyond the typical benchmark examples and considering real applications in astronomy, infectious disease epidemiology, personalised cancer therapy and financial prediction. We anticipate that the emerging success of ABC in producing actual added value and quantitative insights in the real world will continue to inspire a plethora of further applications across different fields of science, social science and technology.
  • Active Learning for Decision-Making from Imbalanced Observational Data
    (2019) Sundin, Iiris; Schulam, Peter; Siivola, Eero; Vehtari, Aki; Saria, Suchi; Kaski, Samuel
     A4 Artikkeli konferenssijulkaisussa
    Machine learning can help personalized decision support by learning models to predict individual treatment effects (ITE). This work studies the reliability of prediction-based decision-making in a task of deciding which action a to take for a target unit after observing its covariates x~ and predicted outcomes p^(y~∣x~,a). An example case is personalized medicine and the decision of which treatment to give to a patient. A common problem when learning these models from observational data is imbalance, that is, difference in treated/control covariate distributions, which is known to increase the upper bound of the expected ITE estimation error. We propose to assess the decision-making reliability by estimating the ITE model’s Type S error rate, which is the probability of the model inferring the sign of the treatment effect wrong. Furthermore, we use the estimated reliability as a criterion for active learning, in order to collect new (possibly expensive) observations, instead of making a forced choice based on unreliable predictions. We demonstrate the effectiveness of this decision-making aware active learning in two decision-making tasks: in simulated data with binary outcomes and in a medical dataset with synthetic and continuous treatment outcomes.
  • Adaptive real-time anomaly detection for multi-dimensional streaming data
    (2017-04-03) Saarinen, Inka
     Perustieteiden korkeakoulu | Master's thesis
    Data volumes are growing at a high speed as data emerges from millions of devices. This brings an increasing need for streaming analytics, processing and analysing the data in a record-by-record manner. In this work a comprehensive literature review on streaming analytics is presented, focusing on detecting anomalous behaviour. Challenges and approaches for streaming analytics are discussed. Different ways of determining and identifying anomalies are shown and a large number of anomaly detection methods for streaming data are presented. Also, existing software platforms and solutions for streaming analytics are presented. Based on the literature survey I chose one method for further investigation, namely Lightweight on-line detector of anomalies (LODA). LODA is designed to detect anomalies in real time from even high-dimensional data. In addition, it is an adaptive method and updates the model on-line. LODA was tested both on synthetic and real data sets. This work shows how to define the parameters used with LODA. I present a couple of improvement ideas to LODA and show that three of them bring important benefits. First, I show a simple addition to handle special cases such that it allows computing an anomaly score for all data points. Second, I show cases where LODA fails due to lack of data preprocessing. I suggest preprocessing schemes for streaming data and show that using them improves the results significantly, and they require only a small subset of the data for determining preprocessing parameters. Third, since LODA only gives anomaly scores, I suggest thresholding techniques to define anomalies. This work shows that the suggested techniques work fairly well compared to theoretical best performance. This makes it possible to use LODA in real streaming analytics situations.
  • AI-assisted curriculum learning
    (2022-07-29) Xiao, Haoping
     Perustieteiden korkeakoulu | Master's thesis
    Deep reinforcement learning is widely applied in de novo molecular design to generate molecules with desired properties. This technique often has a sparse reward problem since the target properties usually exist for the minority of the generated molecules. With a sparse reward, the agent in a de novo design tool may fail to begin learning and waste much time exploring areas in the vast chemical space that are far away from the target area. A recent study successfully applied curriculum learning to mitigate the sparse reward problem. However, a chemist must hand-craft a curriculum for the generative agents, which requires domain knowledge and is time-consuming, especially as tasks grow in complexity. This thesis applies an AI assistance framework to assist in a curriculum design task by recommending actions. The AI assistant infers the private information of the chemist, including design objective function and chemist’s biases. Then, the AI tries to convince the chemist to adopt its advice. The chemist is free to choose action after receiving advice. This setting presents a significant improvement in AI safety. We demonstrate this method with a simulated chemist in a de novo design task, where the generated molecules should be predicted to be active against the dopamine type 2 receptor (DRD2). Our experiments show that the AI-assisted curriculum learning achieves a pronounced improvement on the sparse property (DRD2) and significantly outperforms unassisted curriculum learning.
  • Aktiivisten DNA-muutosten seulonta riippuvuusmalleilla
    (2010) Huovilainen, Olli-Pekka
     Elektroniikan, tietoliikenteen ja automaation tiedekunta | Master's thesis
    Syövän kehittymiseen liittyy geneettiset muutokset useissa solun kasvuun, jakautumiseen tai kuolemaan liittyvissä geeneissä. Näissä syöpään liittyvissä geeneissä mutaatiot aiheuttavat muutoksia geenin aktiivisuudessa syöpäsoluissa. Sekä mutaatioita että geenien aktiivisuuksia voidaan mitata geenisiruilla. Näiden kopioluku- ja ilmentymämittausten avulla voidaan etsiä syöpään liittyviä geenejä. Tässä työssä tutkittiin todennäköisyysperusteiseen kanoniseen korrelaatioanalyysiin perustuvien riippuvuusmallien käyttämistä syöpägeenien etsimisessä. Tässä menetelmässä etsintä tehdään tutkimalla kopioluku- ja ilmentymämittauksien yhteyksiä riippuvuusmalleilla kunkin geenin ympäristössä. Nämä riippuvuusmallit mahdollistavat myös etukäteistiedon hyväksikäytön rajoittamalla tutkittava riippuvuutta. Syöpägeenien etsinnässä voidaan käyttää etukäteistietona syöpägeeneihin liittyvän kopioluku- ja ilmentymämuutoksien paikkariippuvuutta. Tällä rajoitettiin menetelmän etsimä riippuvuus vain saman geenin mittausten välille. Tämä pienensi pienestä näytemäärästä johtuvaa mallin ylisovitusta. Rajoitettujen riippuvuusmallien käyttö paransi menetelmän toimivuutta selkeästi. Menetelmän todettiin toimivan parhaiten sallimalla pieni vapaus rajoitukselle. Työssä toteutettiin avoimen lähdekoodin sovellus syöpägeenien etsimiseen riippuvuusmalleilla. Menetelmän toimivuutta verrattiin muihin ilmentymä- ja kopiolukumittausten riippuvuuksien tutkimiseen tarkoitettuihin menetelmiin. Rajoitettuihin riippuvuusmalleihin perustuvan menetelmän todettiin toimivan paljon paremmin syöpägeenien etsinnässä kuin muut verratut menetelmät. Tässä työssä toteutettu menetelmä on saatujen tulosten perusteella paras menetelmä syöpägeenien etsinnässä kopioluku- ja ilmentymämittauksilla.
  • Analysis of differences between metabolic time series with hidden Markov models
    (2007) Ermolov, Andrey
     Helsinki University of Technology | Master's thesis
    In the thesis the method for finding and analyzing differences between sparse metabolic time series was developed. In metabolic time series the measurements contain concentrations of chemical compounds produced in reactions in a living organism. Analyzing sparse metabolic time series is an important task in medicine and biology, because the metabolome contains a lot of information about the organism, for example about diseases or pathologies, but at the same time it is usually difficult and expensive to make frequent measurements. The most important characteristics of the data used in the study are that time series are relatively short and sparse (that is time interval between subsequential observations is considerably longer than duration of the most biochemical reactions in an organism), measurements are confounded with heavy noise, and the number of time series available is considerably smaller than the dimension of the measurements. The developed approach was primarily designed for metabolomic data, but it can also be applied to the time series with the similar characteristics in other fields. The developed approach contains four stages: preprocessing, designing statistical model, finding differences and analyzing their statistical significance. Hidden Markov Models (HMM) are employed to find differences between metabolic time series. HMM is a statistical method where the modeled system is assumed to be a Markov chain with unknown ("hidden") states emitting visible observations. The properties of the underlying process can be analyzed based on the characteristics of the hidden states and their interrelationships. The developed method was succesfully applied to find and analyze differences between metabolic time series of males and females in growing age extracted from blood plasma. Several time-dependent between-gender differences were identified. Justified suggestions about where these differences come from and about their general structure were made. Compared to methods that ignore the time series structure, HMM-based approach gives superior results and provides some completely new insights to between-gender differences, for instance progression of the development can be investigated. HMMs also combine several advantages compared to other time series modelling methods: they are computationally relatively light, are able to produce relatively good results with the moderate amount of data and can be applied to sparse time series. It is relatively easy to extend and generalize the developed method.
  • Approximate Bayesian Computation with Domain Expert in the Loop
    (2022) Bharti, Ayush; Filstroff, Louis; Kaski, Samuel
     A4 Artikkeli konferenssijulkaisussa
    Approximate Bayesian computation (ABC) is a popular likelihood-free inference method for models with intractable likelihood functions. As ABC methods usually rely on comparing summary statistics of observed and simulated data, the choice of the statistics is crucial. This choice involves a trade-off between loss of information and dimensionality reduction, and is often determined based on domain knowledge. However, handcrafting and selecting suitable statistics is a laborious task involving multiple trial-and-error steps. In this work, we introduce an active learning method for ABC statistics selection which reduces the domain expert’s work considerably. By involving the experts, we are able to handle misspecified models, unlike the existing dimension reduction methods. Moreover, empirical results show better posterior estimates than with existing methods, when the simulation budget is limited.
  • A B2B Buyer’s Purchasing Behavior: An Agent-Based Modeling Approach
    (2019-10-23) Saaranto, Alina
     Perustieteiden korkeakoulu | Master's thesis
  • Balancing Imbalanced Toxicity Models : Using MolBERT with Focal Loss
    (2025) Masood, Muhammad Arslan; Kaski, Samuel; Ceulemans, Hugo; Herman, Dorota; Heinonen, Markus
     A4 Artikkeli konferenssijulkaisussa
    Drug-induced liver injury (DILI) presents a multifaceted challenge, influenced by interconnected biological mechanisms. Current DILI datasets are characterized by small sizes and high imbalance, posing difficulties in learning robust representations and accurate modeling. To address these challenges, we trained a multi-modal multi-task model integrating preclinical histopathologies, biochemistry (blood markers), and clinical DILI-related adverse drug reactions (ADRs). Leveraging pretrained BERT models, we extracted representations covering a broad chemical space, facilitating robust learning in both frozen and fine-tuned settings. To address imbalanced data, we explored weighted Binary Cross-Entropy (w-BCE) and weighted Focal Loss (w-FL) . Our results demonstrate that the frozen BERT model consistently enhances performance across all metrics and modalities with weighted loss functions compared to their non-weighted counterparts. However, the efficacy of fine-tuning BERT varies across modalities, yielding inconclusive results. In summary, the incorporation of BERT features with weighted loss functions demonstrates advantages, while the efficacy of fine-tuning remains uncertain.
  • Batch production optimization by evolutionary computation and predictive simulation
    (2001) Parviainen, Olli
     Helsinki University of Technology | Master's thesis
    Laajojen teollisuusprosessien hallintaa voidaan helpottaa ennakoivalla simuloinnilla, jolloin prosessin käyttäytymistä lähitulevaisuudessa simuloidaan tietokoneen avulla. Ennakoivan simuloinnin avulla pystytään tunnistamaan tulevia ongelmatilanteita etukäteen ja puuttumaan ongelmiin hyvissä ajoin. Työssä tutkittiin ennakoivaan simulointiin perustuvaa yleiskäyttöistä ajosuunnittelumenetelmää, jonka avulla panosprosesseille voitaisiin automaattisesti tuottaa tulevien vuorokausien aikana suoritettavat ohjaustoimenpiteet sisältävä ajosuunnitelma. Menetelmässä prosessikohtaiset ajosuunnittelutavoitteet esitetään optimointitehtävänä, joka ratkaistaan geneettisellä algoritmilla. Geneettisen algoritmin kohdefunktio lasketaan simuloiden prosessin käyttäytymistä suunniteltavan ajanjakson aikana. Ajosuunnittelumenetelmä toimi hyvin, kun sitä sovellettiin kahden erityyppiseen esimerkkiprosessin ajosuunnittelutehtävien ratkaisemiseen. Perinteisesti panosprosessien ajosuunnitteluun on käytetty operaatiotutkimuksen menetelmiä, kuten lineaarista sekalukuoptimointia. Näihin menetelmiin verrattuna työssä käytetyllä menetelmällä voidaan ratkaista myös epälineaarisia ajosuunnittelutehtäviä ja hyödyntää suoraan tuotantolinjan simulointimallia.
  • Bayesian clustering of huge friendship networks
    (2007) Aukia, Janne
     Helsinki University of Technology | Master's thesis
    Sosiaalisten verkkopalveluiden, joita ovat esimerkiksi MySpace, Facebook ja Last.fm, viimeaikaisen suosion kasvun myötä kiinnostus erittäin suurten ystävyysverkostojen analysointiin on kasvanut. Näissä verkoissa on jopa miljoonia solmuja, joten ne tarjoavat hyvän testiympäristön uusille verkkoalgoritmeille. Verkkojen analysointimenetelmiä voidaan hyödyntää myös muihin kuin sosiaalisiin verkkoihin, kuten proteiinien välisiin vuorovaikutusverkkoihin ja verkkosivujen välisiin linkkeihin. Sosiaalisilla verkostoilla on tyypillisesti rakenne: niissä on tiheitä solmuryhmittymiä, ja joillakin solmuilla on suhteettoman paljon yhteyksiä. Rakenne syntyy, koska ystävyydet eivät muodostu satunnaisesti. Ihmiset sen sijaan tapaavat ystävystyi samanlaisten ihmisten kanssa. Tätä voi kutsua homofiliaksi. Ystävyyksien syntyyn vaikuttavat myös muut tekijät kuten maantieteellinen sijainti ja yhteisiin aktiviteetteihin osallistuminen. M0-algoritmi löytää klusterirakenteen homofiilisistä verkoista bayesilaisen tilastollisen inferenssin avulla. Algoritmi pohjautuu generatiiviseen malliin, jossa verkon sivut luodaan latenttien komponenttien perusteella. Mallin parametrien tilastollisessa päättelyssä käytetään Gibbs-otantaa. Homofilian vuoksi samaan klusteriin kuuluvilla solmuilla on todennäköisesti yhteisiä piirteitä. Tässä diplomityössä esitetään MO-algoritmille tehokas toteutus, joka käyttää tasapainotettua binääripuuta komponenttien todennäköisyyksien tallennukseen. Toteutus toimii jopa miljoonien solmujen verkoilla. Algoritmia testataan joukolla aiemmin tutkittuja pieniä verkkoja ja Last.fm-palvelusta kerätyllä ystävyysverkolla, jossa on yli 600 000 käyttäjää. Algoritmi löytää merkityksellisiä rakenteita monenkokoisista verkoista, ja tulokset ovat vertailukelpoisia hierarkisilla klusterointimenetelmillä saatujen tulosten kanssa. Menetelmän vahvuus on solmujen sumea klusterointi, jossa solmu voi kuulua samanaikaisesti useaan klusteriin. Hyperparametrien valinta on kuitenkin usein hankalaa.
  • Bayesian exponential family projections
    (2010) Virtanen, Seppo Juhani
     School of Electrical Engineering | Master's thesis
    Exploratory data analysis stands for extracting useful information from data sets. Machine learning methods automate this process by fitting models to data. It is essential to provide all available background knowledge for building such models. Principal component analysis is a standard method for exploratory data analysis. Recently its probabilistic interpretation has illustrated that it is only suitable for a specific type of data. Extension of principal component analysis to the exponential family removes this problem. In this thesis a general model family suitable for the analysis of multiple data sources is presented by building on the exponential family principal component analysis. The unifying framework contains as special cases methods suitable for unsupervised and supervised learning. While earlier methods have mainly relied on maximum likelihood inference, in this thesis Bayesian modelling is chosen. In Bayesian modelling background knowledge is utilized in the form of prior distributions. In this thesis, a general prior distribution is proposed that takes distribution-specific constraints into account. Multiple contributions to modelling, inference and model interpretation are introduced. With empirical experiments it is demonstrated how the proposed methods outperform traditional methods.
  • Bayesian Inference for Optimal Transport with Stochastic Cost
    (2021) Mallasto, Anton; Heinonen, Markus; Kaski, Samuel
     A4 Artikkeli konferenssijulkaisussa
    In machine learning and computer vision, optimal transport has had significant success in learning generative models and defining metric distances between structured and stochastic data objects, that can be cast as probability measures. The key element of optimal transport is the so called lifting of an exact cost (distance) function, defined on the sample space, to a cost (distance) between probability measures over the sample space. However, in many real life applications the cost is stochastic: e.g., the unpredictable traffic flow affects the cost of transportation between a factory and an outlet. To take this stochasticity into account, we introduce a Bayesian framework for inferring the optimal transport plan distribution induced by the stochastic cost, allowing for a principled way to include prior information and to model the induced stochasticity on the transport plans. Additionally, we tailor an HMC method to sample from the resulting transport plan posterior distribution.
  • Bayesian integrative modelling of metabolic and transcriptional data using pathway information
    (2010) Mohammadi, Pejman
     School of Science | Master's thesis
    One of the rising trends in computational systems biology is to characterize biological systems through integrated analysis of different sources of biological information. While gene expression and metabolic measurements are among the most prevalent biological information sources their integration is problematic due to the difficulties raised by the high dimensionality of the datasets, excessive noise and lack of data samples. The biochemical skeleton of metabolism has been extensively studied an widely used for simulating the metabolic behaviours in cells. Nevertheless, the rigid structure of metabolism can also be utilized as a scaffold for analysis of the genome-scale datasets. This helps to manage the high dimensionality and noise more effectively while also providing a natural link for integrating several omics datasets in the mean time. The ultimate goal of the work presented in this thesis is to develop a novel data fusion scheme by taking advantage of the genome-scale reconstructed models of metabolism as prior knowledge for integrated analysis of transcriptional and metabolic measurements.
  • Bayesian metabolic flux analysis reveals intracellular flux couplings
    (2019-07-15) Heinonen, Markus; Osmala, Maria; Mannerström, Henrik; Wallenius, Janne; Kaski, Samuel; Rousu, Juho; Lähdesmäki, Harri
     A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä
    Motivation: Metabolic flux balance analysis (FBA) is a standard tool in analyzing metabolic reaction rates compatible with measurements, steady-state and the metabolic reaction network stoichiometry. Flux analysis methods commonly place model assumptions on fluxes due to the convenience of formulating the problem as a linear programing model, while many methods do not consider the inherent uncertainty in flux estimates. Results: We introduce a novel paradigm of Bayesian metabolic flux analysis that models the reactions of the whole genome-scale cellular system in probabilistic terms, and can infer the full flux vector distribution of genome-scale metabolic systems based on exchange and intracellular (e.g. 13C) flux measurements, steady-state assumptions, and objective function assumptions. The Bayesian model couples all fluxes jointly together in a simple truncated multivariate posterior distribution, which reveals informative flux couplings. Our model is a plug-in replacement to conventional metabolic balance methods, such as FBA. Our experiments indicate that we can characterize the genome-scale flux covariances, reveal flux couplings, and determine more intracellular unobserved fluxes in Clostridium acetobutylicum from 13C data than flux variability analysis.
  • Bayesian Optimization Augmented with Actively Elicited Expert Knowledge
    (2022-06-13) Huang, Daolang
     Perustieteiden korkeakoulu | Master's thesis
    Bayesian optimization (BO) is a well-established method to optimize black-box functions whose direct evaluations are costly. In this thesis, we tackle the problem of incorporating expert knowledge into BO, with the goal of further accelerating the optimization, which has received very little attention so far. We design a multi-task learning architecture for this task, with the goal of jointly eliciting the expert knowledge and minimizing the objective function. In particular, this allows for the expert knowledge to be transferred into the BO task. We introduce a specific architecture based on Siamese neural networks to handle the knowledge elicitation from pairwise queries. Experiments on various benchmark functions with both simulated and actual human experts show that the proposed method significantly speeds up BO even when the expert knowledge is biased compared to the objective function.
  • Bayesian Optimization for Partially Overlapping Covariate Data Sources
    (2022-06-13) Nguyen, Dan
     Perustieteiden korkeakoulu | Master's thesis
    One problem in the real-world industrial process is how to utilize diverse information on best practices through different data sources. It becomes more complicated when those best practices are different, but not entirely, from each other. The goal is to find the optimal best practices from those diverse and somewhat different data. That problem has been formulated into finding the optimal parameter settings in diverse, partially overlapping covariate data sources. First, the data from different sources are stacked row-wise to form a master data set with missing data. Then, Bayesian Optimization with Missing Inputs is employed to find the optimal experiment's parameter settings. Different methods of modeling the missing data set are tested, such as Bayesian Non-negative Matrix Factorization (BNMF) and Bayesian Probabilistic Matrix Factorization (BPMF). Both provide a quality representation of the missing data, allowing the Bayesian Optimization algorithm to work. The BPMF-based methods have significantly better performances than the BNMF-based methods. However, BNMF-based methods are helpful in some specific cases due to the structure of the missing data set. Multi-armed Bandit Algorithms are used to tackle the problem of a parameter settings budget constraint in each iteration. The $\epsilon$-greedy and UCB1 have been tested. The $\epsilon$-greedy can occasionally give better results because of its randomness. In contrast, The UCB1 consistently improves its performance through each iteration. This work proposes a framework to utilize the information from partially overlapping data sources to find the parameter settings that yield a maximum return. This work benefits a wide range of real-world industrial production processes and opens exciting research directions.
  • Bayesian reduced rank regression for assessment of age and cognition related effects on stable patterns in EEG recordings
    (2022-03-21) Ala-Kulju, Kimmo
     Perustieteiden korkeakoulu | Master's thesis
    Increasing incidence of dementias due to the globally ageing population sets new demands for accessible diagnostics. Widely available electroencephalography (EEG) combined to machine learning techniques has been shown to have utility in early dementia detection. EEG recordings are additionally known to exhibit longitudinally stable properties over a time period of months to years. Widespread degenerative changes in the brain associated to normal ageing and especially to dementia is hypothesized to have an effect on the stability of EEG. This thesis examined age and cognition related effects on the stability of EEG recordings utilizing a probabilistic latent variable model termed Bayesian reduced rank regression (BRRR). The model was applied to power spectral features of EEG recordings from a dataset of 42 healthy subjects with two repeated measurements each. The low-dimensional stable representations obtained by the BRRR model achieved better stability metrics compared to the use of dimension reduction with probabilistic principal component analysis (PPCA) or full data without dimension reduction. Age and cognitive test scores were used as surrogate variables for degenerative changes in the brain, but neither had any clearly observable effect on the stability measures defined as the within-subject distances in the stable subspace. Reconstruction errors for the individual EEG recordings were additionally examined to assess their conformity to the stable correlation structures learnt by the model, but no apparent correlation between the errors and the age or cognitive performance of the subjects was observed. A major data related limitation of the study was the subjects' relatively young age range of 17--71 years and the lack of diagnosed dementia cases, requiring the use of weaker surrogate variables instead. Further studies on dementia and the stability of EEG with more extensive dataset are recommended for more conclusive analysis.
  • Bayesian reduced rank regression models generalizable neural fingerprints that differentiate between individuals in magnetoencephalography data
    (2024-05) Haakana, Joonas; Merz, Susanne; Kaski, Samuel; Renvall, Hanna; Salmelin, Riitta
     A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä
    Recent magnetoencephalography (MEG) studies have reported that functional connectivity (FC) and power spectra can be used as neural fingerprints in differentiating individuals. Such studies have mainly used correlations between measurement sessions to distinguish individuals from each other. However, it has remained unclear whether such correlations might reflect a more generalizable principle of individually distinctive brain patterns. Here, we evaluated a machine-learning based approach, termed latent-noise Bayesian reduced rank regression (BRRR) as a means of modelling individual differences in the resting-state MEG data of the Human Connectome Project (HCP), using FC and power spectra as neural features. First, we verified that BRRR could model and reproduce the differences between metrics that correlation-based fingerprinting yields. We trained BRRR models to distinguish individuals based on data from one measurement and used the models to identify subsequent measurement sessions of those same individuals. The best performing BRRR models, using only 20 spatiospectral components, were able to identify subjects across measurement sessions with over 90% accuracy, approaching the highest correlation-based accuracies. Using cross-validation, we then determined whether that BRRR model could generalize to unseen subjects, successfully classifying the measurement sessions of novel individuals with over 80% accuracy. The results demonstrate that individual neurofunctional differences can be reliably extracted from MEG data with a low-dimensional predictive model and that the model is able to classify novel subjects.
  • Bayesian solutions to the label switching problem
    (2008) Puolamäki, Kai; Kaski, Samuel
     Faculty of Information and Natural Sciences | D4 Julkaistu kehittämis- tai tutkimusraportti taikka -selvitys
    The label switching problem, the unidentifiability of the permutation of clusters or more generally latent variables, makes interpretation of results computed with MCMC sampling difficult. We introduce a fully Bayesian treatment of the permutations which performs better than alternatives. The method can be used to compute summaries of the posterior samples even for nonparametric Bayesian methods, for which no good solutions exist so far. Although being approximative in this case, the results are very promising. The summaries are intuitively appealing: A summarized cluster is defined as a set of points for which the likelihood of being in the same cluster is maximized.