4th International Workshop on Cognitive Information Processing

From Brede Wiki
(Redirected from CIP 2014)
Jump to: navigation, search
Event (help)
4th International Workshop on Cognitive Information Processing (CIP 2014)
Location: Copenhagen Denmark
Date & time: 2014-05-26 – 2014-05-18

International Workshop on Cognitive Information Processing

Databases: Twitter WikiCFP
Link: http://cip2014.conwiz.dk
Search: DuckDuckGo Google Bing Twitter

4th International Workshop on Cognitive Information Processing (CIP 2014) is the 2014 workshop in the workshop series International Workshop on Cognitive Information Processing (CIP) on cognitive information processing.

Program: http://www.conwiz.dk/cgi-all/cip2014/list_program.pl

Lars Kai Hansen was general chair. Keynote by Klaus-Robert Müller ("Multimodal imaging, non-stationarity and brain-computer interfacing"), Søren Bech, Samuel Kaski, Jose C. Principe and invited lectures by Stefania Serafin and Preben Kidmose supported by the CoSound and Neuro24/7 research projects.


[edit] Talks

[edit] Keynote

Klaus-Robert Müller 
"Multimodal imaging, non-stationarity and brain-computer interfacing". Müller has been working with brain-computer interfaces with electroencephalography (EEG). His group has advanced the EEG BCI method with little (<10min) training using left/right imagined movement. From 200 Hz down-sampled filter they extract features based on 1) FFT-based low-pass filter, 2) band-pass 4-40 Hz AR coefficients, 3) subject-specific band-pass filter e.g., 7-14 Hz with multi-class common spatial pattens and perform artifact removal. These features are piped into a classifier. His group has collaborated with Roderick Murray-Smith on a EEG-based spelling. They apply stationary subspace analysis (SSA) to the EEG-data (Finding stationary subspaces in multivariate time series and Finding stationary brain sources in EEG data) In the leading SSA-components Müller finds signal related to loose electrodes, eye movements and muscle activity. "stationarized common spatial patterns". "Divergence framework for Common Spatial Patterns Algorithms" matlab software available from http://www.divergence-methods.org. "Multimodal source power correlation analysis" (mSPoC). Müller has worked with multimodal EEG-NIRS data. SSA splits the signal into a stationary and a non-stationary part:[1]
<math>x(t) = As(t) = [A^s A^n ] \left[ \begin{array}{c} S^s(t) \\ S^n(t) \end{array} \right]</math>
See also http://bbci.de/ The lecture slides available from http://cip2014.conwiz.dk/files/mueller_cip2014.pdf

[edit] Monday sessions

Sebastian Weichwald 
"Decoding index finger position from EEG using random forests". [1]. Subjects should press keys on a numerical keyboard and the task was to decode the finger position based on EEG. The EEG-data with a total of 26819 trials was spatially filtered with a Laplacian. Extracted 8904 features and tested with leave-one-subject-out cross-validation using a random forest classifier.
Elvira Khachatryan 
"Amplitude of N400 component unaffected by lexical priming for moderately constraining sentences". 32-electrode EEG recording with BioSemi downsampled from 2048 to 256 Hz, filtering with 4th order Butterworth to 0.1 to 30 Hz, removal of eye artifacts, extracting data from 100 ms pre-onset to 1000 ms post-onset
Per Bækgaard
"In the twinkling of an eye: synchronization of EEG and eye tracking based on blink signatures": Using blinks in EEG to synchronize with eye tracking. Spontaneous closing eye: 50-70 ms while associated reopening +100 ms. There is a nice correspondence between eye blink extracted from EEG and the eye tracking systems.
Søren Bech 
"Sensory analysis - a research paradigm for linking perception and physics/signal processing". Stevens' power law. Two ways to identification for attributes: Direct elicitation method consensus vocabulary method ("RaPID", ...), individual vocabulary methods ("free-choice profiling", "repertory grid", "flash profile"). "Perceptually Optimised Sound Zones": http://iosr.surrey.ac.uk/projects/POSZ/
Bob L. Sturm 
"A closer look at deep learning neural networks with low-level spectral periodicity features" [2]. "The Latin music database": "3229 full songs singly labeld in 10 classes" (Axé, Bachata, Forró, Gaúcha, Salsa). "DeSPerF" used of the BALLROOM music dataset[2] Sturm investigates why deep learning neural network perform so well in music genre classification. He finds that the classifier is apparently critically dependent on the tempo. The BALLROOM dataset has a high statistical dependency between genre and tempo.
Hans-Peter Gasselseder 
"Those who played were listening to the music? Immersion and dynamic music in the ludonarrative". Subjective measures: EMuJoy by Nagel et al. 2007, iGEQ, MEC-SPQ, ...
Stefania Serafin 
"Sonic Interaction Design in Experimental Music, Virtual Reality, Cultural Heritage and the Automotive Industry". She has worked with "augmented threadmill" with visual-audio-haptic walking. Virtual walking: mechanical repositioning walking, (illusionary walking), walking-in-place walking. With inspiration from the parchment-skin illusion manipulated the sound associated with a pulley machine. Reconstruction with physical modeling of old audio generated instruments: Intonarumori and DREAM.
Paul Kendrick 
Automatic detection of microphone handling noise. He did a web survey on most commonly reported recording errors: Background noise, wind-induced noise, handling noise. He wants to make a machine learning model to detect, e.g., handling noise. On sound recordings he extract MFCCs and classify on whether there are handling noise in the recording with a random forest classifier. C++ code for his project is available at http://github.com/kenders2000/WindNoiseDetection An app for wind noise has also been constructed.
David Meredith 
Compression-based geometric pattern discovery in music. "maximal translatable patterns" (MTP). COSIATEC algorithms to find interesting repeated patterns in music. http://www.liederenbank.nl
Jesper Steen Andersen 
Using the Echo Nest's automatically extracted music features for a musicological purpose. Looked on labelings from Echo Nest. It labels with energy, liveness, speechiness, acousticness, danceability, valence.

[edit] Tuesday sessions

Samuel Kaski 
Exploratory and contextual search with interactive intent modelling. Three different topics presented: 1) Intent Radar. Relevant paper: Directing exploratory search with interactive intent modeling [3]. "Intent Radar" is an interactive radial layout of keywords. Example on SciNet data. Keywords are layout with radius according to relevance and angle according to similarity. Optimize the angle by MDS NeRV(???). http://augmentedresearch.hiit.fi YouTube video on the system: https://www.youtube.com/watch?v=zOoFNpF6eFk 2) Eye tracking and EEG in information retrieval. Manuel Eugster et al 2014 Predicting term-relevance from brain signals. 3) "Intelligent Information Access": Augmented reality using eyetracking, face recognition, information retrieval and eye glass display. Ajanki et al. Virtual Reality, 2011. https://www.youtube.com/watch?v=gtuGSWDVdQU EU-project http://www.mindsee.eu/
Will Penny 
Simultaneous localisation and planning [4]. Introduction: Mumford (1992): "The Bayesian brain" with message passing. "Sharp wave ripples" in hippocampus. Paper: Forward and backward inference in spatial cognition. entorhinal as forward inference. A Markov model of CA3/CA1: a hidden Markov model with two different source of information: sensory and goals (dopaminergic). Maze represented in a matrix indicating the probability/possibility(?) of going from one cell to another. This is inspired from work of Kappen, PRL 2005 and Emanuel Todorov NIPS 2006 (Linearly-solvable Markov decision problems)

This is a graph with borders and nodes. Maybe there is an Imagemap used so the nodes may be linking to some Pages.
Uwe Mönks 
Fast evidence-based information fusion
Francesco A. N. Palmieri 
Belief propagation and learning in convolution multi-layer factor graphs. "Normal graph" (Codes on graphs: normal realizations). Related papers Simulink implementation of belief propagation in normal factor graphs.

Round table discussion with Will Penny, Jose Principe, Sergios Theodoridis, Anibal Figueiras Vidal and Samuel Kaski. Topics: "Big data", online learning, distributed data, scalability. Life logging helping people with dementia. "Everything that can be automated will be automated". "The rush to big data is flawed". Experimental design is important. User modeling, distributed learning, distributed systems, topology, "Big Ideas", Bayesian inference, ...

Jose Principe 
A Cognitive Architecture for Object Recognition in Video. Data from a number of diferent image and audio databases, e.g., University of Iowa Musical Instrument notes.
(Francesco Castaldo), Francesco A. N. Palmieri 
Application of factor graphs to multi-camera fusion for maritime tracking.
Pantelis Bouboulis 
Robust Image Denoising In RKHS Via Orthogonal Matching Pursuit <math>y_i = \hat{f}(x_i) = f(x_i) + \eta_i + u_i</math>, where ui is to model the outliers which are sparse. Minimize ||u||_0 subject to a constraint. Their method, KROMP, compared with the BM3D method on the Lena image.
Chong Wang 
A new hand gesture recognition algorithm based on joint color-depth superpixel earth mover's distance. Uses the Kinect that has built-in human body skeleton tracking. Demonstration on a paper-scissor-stone-like game and manipulation of a 3D object in a 3D object viewer.
Isabel Valera, Francisco J. r. Ruiz, Fernando Perez-Cruz 
Infinite factorial unbounded hidden Markov model for blind multiuser channel estimation. Unknown channel length and unknown number of transmitters. Bayesian non-parametrics (BNP). Infinite factorial unbounded-state hidden Markov model.

[edit] Wednesday session

Preben Kidmose og Danilo P. Mandic 
Ear-EEG - a novel brain monitoring methodology. Related papers: A study of evoked potentials from ear-EEG and Auditory evoked responses from ear-EEG recordings.
Camilla Birgitte Falk Jensen 
Emotional responses as independent components In EEG. Common affective ERP components: P100/N100, EPN, P300, LPP. The experiment involved 4 male subject looking on IAPS (pleasant/unpleasant/neutral) images and recording EEG with a BioSemi system, and analysis with, e.g., independent component analysis.
(Jair Montoya-Martínez), Antonio Artés-Rodríguez, (Massimiliano Pontil
Structured sparse-low rank matrix factorization for the EEG inverse problem. EEG source reconstruction.<math>\mathbf{Y = AS+E}</math>. Cost function with either, minium normm e, minimum current estimate (LASSO), mixed norm estimate (group LASSO), sparse group LASSO. Here a matrix factorization approach <math>argmin_{B,C} \left\{ 1/2 || A(BC) - Y||^2_F + \lambda||B||_{2,1} + 1/2||C||^2_F, \lambda >0 \right\}</math>, where B is a "sparse coding matrix" and C "contains the activeity of the latex sources". Alaternating minimization algorithm solved B and C separately, where B is solved with the FISTA method and C directly.
Pietro Stinco 
Channel parameters estimation for cognitive radar systems
(Bijit Kumar Das), Jeronimo Arenas Garcia 
A comparative study of two popular families of sparsity-aware adaptive filters. Sparse systems could be for network echo cancellation or acoustic echo cancellation. Two methods to identify the sparse system: Proportionate-type NLMS (PtNLMS) and sparsity-norm regularized adaptive filters (S-AF). ZA-LMS Gu 2009 zero-attracting. See also l0 Norm Constraint LMS Algorithm for Sparse System Identification
(Mikkel N. Schmidt), (Tue Herlau), Morten Mørup 
Discovering hierarchical structure in normal relational data. "Direct similarity" and "structural similarity". Prior over trees (multifurcating trees): Nested Chinese restaurant process / Gibbs fragmentation trees. Inspired by Marcus Raichle: The restless brain.
Luis Muñoz-González, Miguel Lázaro-Gredilla, Aníbal Ramón Figueiras-Vidal 
Laplace approximation with Gaussian processes for volatility forecasting. For volatility forecasting GARCH models are commonly used. Here a Gaussian process for regression (GPR) is used. Compared withe Girolami & Calderhead 2011 synthetic dataset.

[edit] Papers

  1. Decoding index finger position from EEG using random forests
  2. Amplitude of N400 component unaffected by lexical priming for moderately constraining sentences
  3. In the twinkling of an eye: synchronization of EEG and eye tracking based on blink signatures
  4. Automatic detection of microphone handling noise
  5. Simultaneous localisation and planning
  6. Emotional responses as independent components In EEG
  7. Structured sparse-low rank matrix factorization for the EEG inverse problem
  8. Discovering hierarchical structure in normal relational data

[edit] References

  1. Finding stationary subspaces in multivariate time series
  2. Evaluating rhythmic descriptors musical genre classification
Personal tools