The main obstacle to weakly supervised semantic image segmentation is the difficulty of obtaining pixel-level information from coarse image-level annotations. Most methods based on image-level annotations use localization maps obtained from the classifier, but these only focus on the small discriminative parts of objects and do not capture precise boundaries. FickleNet explores diverse combinations of locations on feature maps created by generic deep neural networks. It selects hidden units randomly and then uses them to obtain activation scores for image classification. Fick-leNet implicitly learns the coherence of each location in the feature maps, resulting in a localization map which identifies both discriminative and other parts of objects. The ensemble effects are obtained from a single network by selecting random hidden unit pairs, which means that a variety of localization maps are generated from a single image. Our approach does not require any additional training steps and only adds a simple layer to a standard convolutional neural network; nevertheless it outperforms recent comparable techniques on the Pascal VOC 2012 benchmark in both weakly and semi-supervised settings.

A new process for the removal of nitrogen from wastewater is introduced. The process involves three steps: (1) partial nitrification of NH 4 + to NO 2 À ; (2) partial anoxic reduction of NO 2 À to N 2 O; and (3) N 2 O conversion to N 2 with energy recovery by either catalytic decomposition to N 2 and O 2 or use of N 2 O to oxidize biogas CH 4 . Steps 1 and 3 have been previously established at full-scale. Accordingly, bench-scale experiments focused on step 2. Two strategies were evaluated and found to be effective: in the first, Fe(II) was used to abiotically reduce NO 2 À to N 2 O; in the second, COD stored as polyhydroxybutyrate (PHB) was used as the electron donor for partial heterotrophic reduction of NO 2 À to N 2 O. For abiotic reduction with Fe(II), the efficiency of conversion of NO 2 À to N 2 O was over 90% with 98% nitrogen removal from water. For partial heterotrophic denitrification, different selection conditions were imposed on acetate-and nitrite-fed communities initially derived from waste activated sludge. No N 2 O was detected when acetate and nitrite were supplied continuously, but N 2 O was produced when acetate and nitrite were added as pulses. N 2 O conversionefficiency was dependent upon the method of addition of acetate and nitrite. When acetate and nitrite were added together (coupled feeding), the N 2 O conversion efficiency was 9-12%, but when acetate and nitrite additions were decoupled, the N 2 O conversion efficiency was 60-65%. Decoupled substrate addition selected for a microbial community that accumulated polyhydroxybutyrate (PHB) during an anaerobic period after acetate addition then consumed PHB and reduced NO 2 À during the subsequent anoxic period. The biological N removal efficiency from the water was 98% over more than 200 cycles. This indicates that decoupled operation can sustain significant long-term N 2 O production. Compared to conventional nitrogen removal, the three-step process, referred to here as Coupled Aerobic-anoxic Nitrous Decomposition Operation (CANDO), is expected to decrease oxygen requirements, decrease biomass production, increase organic matter available for recovery as biogas methane, and enable energy recovery from nitrogen, but pilot-scale studies are needed. Broader contextThe release of reactive forms of nitrogen is a major environmental threat causing hypoxia and eutrophic zones in water bodies. Globally, rising energy costs and increasingly stringent discharge regulation are major drivers for efficient wastewater treatment processes that lower costs and increase recoverable energy from waste. While many processes recover energy from carbon waste as CH 4 , none recovers energy from waste nitrogen. This work introduces a new wastewater treatment process that removes and recovers energy from nitrogen waste by exploiting the thermodynamic properties of N 2 O for energy recovery. The proposed process, referred to here as Coupled Aerobic-anoxic Nitrous Decomposition Operation (CANDO), involves three steps: (1) partial aerobic nitrication of NH 4 + to ...

When a deep neural network is trained on data with only image-level labeling, the regions activated in each image tend to identify only a small region of the target object. We propose a method of using videos automatically harvested from the web to identify a larger region of the target object by using temporal information, which is not present in the static image. The temporal variations in a video allow different regions of the target object to be activated. We obtain an activated region in each frame of a video, and then aggregate the regions from successive frames into a single image, using a warping technique based on optical flow. The resulting localization maps cover more of the target object, and can then be used as proxy ground-truth to train a segmentation network. This simple approach outperforms existing methods under the same level of supervision, and even approaches relying on extra annotations. Based on VGG-16 and ResNet 101 backbones, our method achieves the mIoU of 65.0 and 67.4, respectively, on PASCAL VOC 2012 test images, which represents a new state-of-the-art.

From May to July 2015, there was a nation-wide outbreak of Middle East respiratory syndrome (MERS) in Korea. MERS is caused by MERS-CoV, an enveloped, positive-sense, single-stranded RNA virus belonging to the family Coronaviridae. Despite expert opinions that the danger of MERS might be exaggerated, there was an overreaction by the public according to the Korean mass media, which led to a noticeable reduction in social and economic activities during the outbreak. To explain this phenomenon, we presumed that machine learning-based analysis of media outlets would be helpful and collected a number of Korean mass media articles and short-text comments produced during the 10-week outbreak. To process and analyze the collected data (over 86 million words in total) effectively, we created a methodology composed of machine-learning and information-theoretic approaches. Our proposal included techniques for extracting emotions from emoticons and Internet slang, which allowed us to significantly (approximately 73%) increase the number of emotion-bearing texts needed for robust sentiment analysis of social media. As a result, we discovered a plausible explanation for the public overreaction to MERS in terms of the interplay between the disease, mass media, and public emotions.

No abstract