University of Tehran Machine Learning and Computational Modeling (MLCM) Lab
Lab Supervisor: Babak N. Araabi
University of Tehran
  Home     People     Research     Publication      Datasets      Events      Contact

End-Stopped Cells Can Link Border Ownership and Perceptual Completion

Psychophysical studies suggest that illusory contours (ICs) emerge due to the mechanism of perceiving depth in the visual cortex. The neuronal mechanisms for this process are not well understood and it is not known how it depends on depth perception. We asked whether border ownership coding (as the mechanism of depth ordering and figure-ground segmentation) could explain the neural structure of ICs. We use a recurrent model of border ownership neurons in V2 and V4. The input provided by simple and end-stopped cells in V1. Neurons in the higher cortical area interact based on correlation-dependent plasticity. The results show the consistency of border ownership assignment with natural scene statistics. The results also suggest ICs that emerged from different inducers are due to the same mechanism that is the interaction of end-stopped cells in the border ownership framework. The proposed mechanism is consistent with spatial and temporal properties of ICs.
Habib Zafarian

 

Ignorance Modeling in Group Decision Making with Dempster-Shafer Theory of Evidence

Photo from "Neural Representation of Subjective Value Under Risk and Ambiguity" by Ifat Levy et al

Risk and ambiguity are two conditions in which the consequences of possible outcomes are not certain. Under risk, the probabilities of different outcomes can be estimated, whereas under ambiguity, even these probabilities are not known. Most people are averse to both of these conditions.
Subject's risk and ambiguity aversion characteristics and estimation of upper and lower bounds of probability is the main parameters for modeling decision making. Dempster-Shafer Theory help us to make decision model. By using this model we can combine member' decision and predict group decision to study on social influence in decision making for ambiguous choices.
Amir Hossein Tehrani-safa

 

A Bio-Physically Plausible Model for Matching Behavior in Dynamic Environment

Previous studies have shown that non-human primates can generate highly stochastic choice behaviour, especially when this is required during a competitive interaction with another agent. To understand the neural mechanism of such dynamic choice behaviour, we propose a biologically plausible model of decision making endowed with synaptic plasticity that follows a reward-dependent stochastic Hebbian learning rule. This model constitutes a biophysical implementation of reinforcement learning, and it reproduces salient features of behavioural data from an experiment with monkeys playing a matching pennies game. Due to interaction with an opponent and learning dynamics, the model generates quasi-random behaviour robustly in spite of intrinsic biases. Furthermore, non-random choice behaviour can also emerge when the model plays against a non-interactive opponent, as observed in the monkey experiment. Now we combined a meta-learning algorithm to our model that accounts for the slow drift in the animal’s strategy based on a process of reward maximization.
Hossein Rafipoor

 

Offline Signature Verification:

Hand-written signature is a well-known behavioral characteristic which is currently used as an authentication method to verify people. The verification procedure is done by Forensic Handwritten Experts (FHEs) which is a time consuming job. So that an automatic signature verification system can accelerate and even relieve mistakes from this well-accepted biometric process.
Automatic signature verification is a complicated system because signature is not a completely repetitive pattern. Moreover, although genuine ones obey some rules, training a system is hard due to sample deficiency whether genuine (positive) or forged (negative) ones.
In the literature, signature is divided into offline and online type. Offline type is just images but online one has extra information such as temporal, speed and pressure information. Our interest is offline signature verification since it is more natural, more applicable and more challenging.
We currently work on Persian offline signature verification. For this manner we made University of Tehran Persian offline signature dataset (UTSig) which is accessible from here.
Amir Soleimani Bajestani

 

Former Researches ...

Writer-dependent Recognition of Farsi Subwords with Learning Capability

This thesis concerns with the recognition of offline Farsi/Arabic handwriting. The overall appearance of each subword in Farsi/Arabic script is described by its shape contour that provides us with a rich set of discriminative characteristics. Our approach is writer-dependent, that is, the system is trained to recognize the subwords written by a particular writer. A fast contour alignment is the central part of the proposed algorithm, where the alignment is performed based on a handful of feature points. An efficient lexicon reduction algorithm based on characteristic loci feature, which works directly on subwords' binary images is proposed as well. Fast and precise alignment along with efficient lexicon reduction and appropriate similarity matching yields a high recognition rate, while keeps the speed high. Our experiment on IBN SINA database shows that the correct classification rate could be as high as 91.08% which is almost 5% higher than the best reported results by other researchers. This figure is achieved merely by subword shape matching, without dots and diacritics, and without any statistical language model.
The generation of Farsi/Arabic handwritten patterns for the application of writer-dependent subword recognition is also considered in this thesis. To learn the writer's script style we use a limited set of subwords called ``basic subwords''. The basic subwords are acquired from a writer using tabular sheets. The ``glyphs'' (simple character or ligature) of the writer's script are extracted from these basic subwords. Using glyphs, we can concatenate them to generate any desirable subword pattern synthetically. Since each glyph can have alternate pattern extracted from other basic subwords, we may have more than one sample for each generated subword. The results indicate that the classification process can leads to an acceptable classification rates via synthesized subwords as reference patterns. The experiments indicate that the subword recognition rate improves up to 10% in comparison with using natural data and up to 6% in comparison with using extended data via perturbation.
Kazim Fouladi

 

A Molecular Dynamic Approach Based on Knowledge-based Force Function for Prediction of Protein Structure

Understanding of how a polypeptide is able to fold to its native structure is one of the central problems in molecular biology. Molecular dynamics simulations have become a powerful technique to study protein folding problem. In this technique, physical energy functions are widely used –which are based on the fundamental analysis of the forces between the particles. The main disadvantage of these functions is that they require substantial computational resources. On the other hand, current force fields are not accurate enough to be able to fold a protein on a computer. In this study, we formulated a force field which is obtained from statistical contact preferences within the known protein structures. An estimate of the inter-atomic forces between any two atoms are calculated using this force field. Since it is statistically derived from known protein structures, we expect better result of finding native state in comparison with physical energy functions. In order to simulate evolution of the proteins using this pairwise potential, we developed an MD tool and then analyzed the results in a continuous space,. The goal is to find a method for protein structure prediction. We hope that this method at least can be a guide to identify topological class of a given protein
Ali Ghaffaari

 

 

Control and Intelligent Processing Center of Excellence, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran