This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision | ||
|
news:20250121 [2025/01/21 14:06] ntuicibmcr |
news:20250121 [2025/02/04 11:50] (current) ntuicibmcr |
||
|---|---|---|---|
| Line 3: | Line 3: | ||
| 演講人: Professor Fushing Hsieh, Department of Statistics, University of California, Davis \\ | 演講人: Professor Fushing Hsieh, Department of Statistics, University of California, Davis \\ | ||
| - | 日期: 2/19上午9:00 ; 3/12上午9:00 ; 3/26上午9:00 ; 4/9上午9:00 \\ | + | 日期: 2/19上午9:00 ; 3/13上午9:00 ; 3/26上午9:00 ; 4/9上午9:00 \\ |
| 方式: 線上; 50min for lecture and flexible 10~30min for Q&A . \\ | 方式: 線上; 50min for lecture and flexible 10~30min for Q&A . \\ | ||
| - | 報名網址: https://docs.google.com/forms/d/e/1FAIpQLSc6u7QYdLSZuVBLbu3-S1L_lxnkv6aQR4N5AXgPeCOZTDT1Rw/viewform?usp=sharing \\ | + | 報名網址:\\ |
| + | https://docs.google.com/forms/d/e/1FAIpQLSc6u7QYdLSZuVBLbu3-S1L_lxnkv6aQR4N5AXgPeCOZTDT1Rw/viewform?usp=sharing \\ | ||
| 報名成功者於活動前一週內以e-mail通知，並提供線上會議連結。\\ | 報名成功者於活動前一週內以e-mail通知，並提供線上會議連結。\\ | ||
| - | 摘要:\\ | + | 摘要:請參考如下連結說明\\ |
| + | https://docs.google.com/document/d/1JZmekrOYxgkg1chbPQVZOezoKyS-ittq/edit?usp=sharing&ouid=114702088670815404083&rtpof=true&sd=true \\ | ||
| - | I.(時間:2/19上午9:00)\\ | ||
| - | |||
| - | We set the scientific goal of data analysis as: Discovering all possible authentic associative relations from X to Y contained in observed data set. To achieve this goal in all sciences, we need to a universal platform for scientists to read, to see and to explain what such relational patterns are. This platform must display such associative relations in a collective fashion through all data-points in the ensemble of {X } and all selected linkages {X Y }. In this 1st topic, we discuss in detail why a model with a known functional structure, such as Y=F(X)+e, is not suit well for our scientific goal. Since such a model needs a lot of ad hoc choices, which could not pass the “test of experience”, which is one key criterion for being qualified as a scientific approach. We propose an approach for constructing such a platform. This construction simply points to unscientific nature of statistical analysis and Machine Learning.\\ | ||
| - | |||
| - | |||
| - | II.(時間:3/12上午9:00) \\ | ||
| - | |||
| - | When a complex system is under study, multiple dimensionalities in X and in Y are common in all sciences. And the essential, but mysterious relations from X to Y might be hiding among interacting effects of various orders among the covariate features {x_1, …, x_K} in X. How to explore such potentials and then successfully capture such relations? This question is especially critical when X is defined by functional or time series. On the other hand, the setting of multiple dimensionality of Y is of particular importance in science. How to properly represent original complex system’s dynamics from X to Y is critically depending on how the multiple dimensionality of Y is handled. We propose to an approach to construct such a representation by discovering essential motifs of Y. We illustrate this topic with a simple hyperspectrum data example: X consisting of functional curves and Y being in R^3 for constructing such a complex system representation.\\ | ||
| - | |||
| - | III.(時間:3/26上午9:00)\\ | ||
| - | |||
| - | The task of deciding a proper representation for the targeted complex system under study is performed by first discovering essential motifs in Y and then fusing such motifs into a 1D categorical response variable "Y" ̂. As such the response-vs-covariate (Re-Co) dynamics underlying X to "Y" ̂ is of classification nature. The collection of categories of "Y" ̂ is naturally imbedded with a hierarchy. Upon this response hierarchical structure, multiple subsystems "Y" ̂_lhave high potentials to give rise to essential and critical information of classification. Such information strangely and surprisingly is derived from two kinds of outlier detection tasks: geometric and information. We argue that the Information Content in Data (ICiD) is achieved by collecting all information contents from all subsystems "Y" ̂_l and full system "Y" ̂. This ICiD is the “brain” of a Classification Robot.\\ | ||
| - | |||
| - | IV. (時間:4/9上午9:00)\\ | ||
| - | In many sciences, X and Y are automatically revealed within the original data set. Since the original data formats could take many diverse forms, such as time series, image, ..etc. None of such formats are suitable for associative pattern evaluations directly. So we need to define and extract X or Y from the original data formats. Again, the choices of X and Y must satisfy the scientific criterion: The Re-Co dynamics from X to Y is an informative representation of the underlying dynamics of complex system under study. This issue of effectively finding an informative representation could be rather difficult. To see and feel such a kind of difficulty, we will go through a series of scientific problems centering around the theme of color in image data. This series of examples once again reiterate that doing science needs human intelligence and creativity among many other things. This is the real starting point of Scientific Data Analysis (SDA).\\ | ||
