- Hydrologic modelling of suspended sediment in a river basin using soft computing and statistical techniques
Anil Kumar, Anurag Malik, G.B.Pant University of Agriculture and Technology, India
In this study, the soft computing techniques such as Artificial Neural Network (ANN) in the form of Co-Active Neuro-Fuzzy Inference System (CANFIS) and Multi-Layer Perceptron (MLP), along with statistical techniques of Multiple Linear and Non-Linear Regressions (MLR and MNLR) was used for simulating the daily SSC at Tekra site of Pranhita River, a major tributary of Godavari River basin in India. The daily data of stream flow and SSC from June, 2000 to November, 2003 were used for calibration and validation of the models used in this study. The combination of appropriate input variables for developing various models was decided using the Gamma Test (GT). The results obtained by CANFIS, MLP, MLR, and MNLR models were compared with the observed SSC on the basis of statistical indices. The CANFIS model performed superior to the MLP, MLR and MNLR models in simulating current day’s SSC for Pranhita River in India.
- Narrow-Angled Traffic Monitoring Heuristics for Acquiring the Complete Image of a Trailer
Takaya Kawakatsu1, Kenro Aihara2, Atsuhiro Takasu2 and Jun Adachi2, 1The University of Tokyo, Japan, 2National Institute of Informatics, Japan
Traffic surveillance cameras mounted on a bridge are used to monitor heavy trailers passing the bridge in chronological order. That can be utilized to investigate any age-related deterioration of the bridge by estimating the bridge's vibration response. Because options for the camera's an- gle and installation point are limited because of privacy protection, we must reconstruct the full out-of-angle scene from pinhole angled movies. In this paper, we utilize a traditional background- subtraction-based tracking algorithm and deep-neural-network-based object separation in our pro- posal for a traffic monitoring system. The background subtraction compensates for the neural network's poor ability to locate objects exactly and can achieve a reasonable reconstruction of the trailer images.
- Weakly Supervised Wildlife Recognition Using Subtitles And Discretized Cnn Features
Aparna Nurani Venkitasubramanian, Tinne Tuytelaars and Marie-Francine Moens, KU Leuven, Belgium
We propose a feature transformation targeted at object recognition in the wild. This feature transformation involves discretization of activations of a convolutional neural network along their dimensions. The resulting bag-of-activations representation allows isolation of key components in the image without the need for an object detection or image segmentation step. We apply this representation for multi-label classification in a multi-modal setting and show that the discretization used with a multinomial Naive Bayes classifier yields much better performance compared to raw features with a traditional Naive Bayes classifier- the precision is more than doubled and the recall is boosted by more than 20% absolute for the task of identifying animals on a challenging dataset of wildlife documentaries with subtitles. The methods proposed here take us a step closer to automatic video indexing.
- Is Ai In Jeopardy? The Need To Under Promise And Over Deliver-The Case For Really Useful Machine Learning.
Martin Ciupa, Calvary Robotics Group, USA
There has been a dramatic increase in media interest in Artificial Intelligence (AI), in particular with regards to the promises and potential pitfalls of ongoing research, development and deployments. Recent news of success and failures are discussed. The existential opportunities and threats of extreme goals of AI (expressed in terms of Superintelligence impacts) are examined with regards to this media "frenzy", and some comment and analysis provided. A recommendation to avoid the extremes of the optimistic and pessimistic positions is offered regarding more realistic short term goals of achieving "Really Useful Machine Learning (RUMLTM). This point is proposed in the context of resolving some of the known issues in bottom-up Deep Learning by Neural Networks, recognized by DARPA as the "Third Wave of AI." An extensive internet accessible reference set of recent supporting media articles is provided.