Exploiting symmetries for scaling loopy belief propagation and relational training

Judging by the increasing impact of machine learning on large-scale data analysis in the last decade, one can anticipate a substantial growth in diversity of the machine learning applications for “big data” over the next decade. This exciting new opportunity, however, also raises many challenges. One of them is scaling inference within and training of graphical models. Typical ways to address this scaling issue are inference by approximate message passing, stochastic gradients, and MapReduce, among others. Often, we encounter inference and training problems with symmetries and redundancies in the graph structure. A prominent example are relational models that capture complexity. Exploiting these symmetries, however, has not been considered for scaling yet. In this paper, we show that inference and training can indeed benefit from exploiting symmetries. Specifically, we show that (loopy) belief propagation (BP) can be lifted. That is, a model is compressed by grouping nodes together that send and receive identical messages so that a modified BP running on the lifted graph yields the same marginals as BP on the original one, but often in a fraction of time. By establishing a link between lifting and radix sort, we show that lifting is MapReduce-able. Still, in many if not most situations training relational models will not benefit from this (scalable) lifting: symmetries within models easily break since variables become correlated by virtue of depending asymmetrically on evidence. An appealing idea for such situations is to train and recombine local models. This breaks long-range dependencies and allows to exploit lifting within and across the local training tasks. Moreover, it naturally paves the way for the first scalable lifted training approaches based on stochastic gradients, both in an online and a MapReduced fashion. On several datasets, the online training, for instance, converges to the same quality solution over an order of magnitude faster, simply because it starts optimizing long before having seen the entire mega-example even once.     

On the communication of scientific data: The Full-Metadata Format

In this paper, we introduce a scientific format for text-based data files, which facilitates storing and communicating tabular data sets. The so-called Full-Metadata Format builds on the widely used INI-standard and is based on four principles: readable self-documentation, flexible structure, fail-safe compatibility, and searchability. As a consequence, all metadata required to interpret the tabular data are stored in the same file, allowing for the automated generation of publication-ready tables and graphs and the semantic searchability of data file collections. The Full-Metadata Format is introduced on the basis of three comprehensive examples. The complete format and syntax are given in the appendix.     

Effects of polyunsaturated fatty acids and their n-6 hydroperoxides on growth of five malignant cell lines and the significance of culture media

We examined effects of polyunsaturated fatty acids (PUFA), their corresponding hydroperoxy fatty acids (hp-PUFA), as well as various pro- and antioxidants on the growth of tumor cells in culture. When cultured in RPMI 1640 medium, A-427 and WEHI clone 13 cells were both highly sensitive to hydroperoxy docosahexaenoic acid (hp-DHA), but they were far less sensitive in minimum essential medium (MEM). In contrast, A-427 cells were also sensitive to DHA in both culture media, while WEHI clone 13 cells, as well as other cell lines, tested in their respective media, were resistant. The lower sensitivity of the cell lines to hp-DHA in MEM-medium was apparently due to a more rapid reduction of hp-DHA to the corre-sponding hydroxy-DHA in MEM-medium. Addition of glutathione (GSH) to the culture medium abolished the effects of hp-DHA, but not the effects of DHA, while depletion of intracellular GSH levels by L-buthionine-S,R-sulfoximine strongly enhanced the cytotoxic effect of hp-DHA, but not the cytotoxic effect of DHA. α-Tocopherol protected A-427 cells against the toxic effect of DHA and abolished the induced lipid peroxidation, while it did not protect against the toxic effects of hp-DHA in A-427 or WEHI clone 13 cells. Ascorbic acid reduced the cytotoxic effect of DHA, but potentiated the toxic effect of hp-DHA while selenite essentially abolished the toxicity of both DHA and hp-DHA. These results indicate that sensitivity of tumor cell lines to PUFA and their oxidation products depends on their antioxidant defense mechanisms, as well as culture conditions, and establishes hp-DHA as a major, but probably not the sole, metabolite responsible for cytotoxicity of DHA.     

Gravitational drainage of compressible organic materials

A model was developed to simulate drainage of compressible particle suspensions, and study how cake compression and volumetric load influence the process. The input parameters were settling velocity, cake resistance and compressibility. These parameters were found using a new experimental method. Dextran‐MnO₂ particle suspensions were drained as these resemble organic waste slurries with respect to settling and compressibility. It was demonstrated that cake compressibility must be taken into account to obtain adequate simulations. This implies that pressurized filtration resistances cannot be used for drainage simulations. In the filtration step, a distinct increase of dry matter from top to bottom of the cake was observed. During the subsequent consolidation, the cake compressed and a uniform dry matter profile was found. The final dry matter content of the cake increased with feed concentration and volumetric load. The drainage time increased proportionally with feed concentration and, more importantly, proportionally with squared volumetric load. © 2010 American Institute of Chemical Engineers AIChE J, 2010