Theses and Dissertations

Permanent URI for this collectionhttp://ir.daiict.ac.in/handle/123456789/1

Browse

Subject: search.filters.subject.Neural Network

Search Results

Now showing 1 - 5 of 5
  • Thumbnail Image
    ItemOpen Access
    Graph Neural Network Based Semantic Mapping and Classification of Dataset for Robotics Applications
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2023) Sharma, Devesh; Jat, P M; Maiti, Tapas Kumar
    In the field of Robotics, deriving meaningful insights from spatial information ispivotal. Our objective was to work with 2D dataset as 3D datasets are relativelymore time consuming as our target is to objective was to do the foundation workfor real time inferences where speed is also an important key factor and so to getbest possible results, trying to improvise in accuracy and speed. Our focus liesin semantic mapping, semantic place classification where mobile robots interpretpartial, noisy sensory data. We delve into end-to-end techniques rooted in probabilisticdeep networks, studying Local-SPNs, Graph SPNs, and TopoNets. Additionally,we explore GNNs for semantic place classification. Our report providessuccinct insights into these methods, emphasizing their principles and implementations,including the local SPN model and GNN for semantic classification. Wegot our best accuracy with GNN 70.15% which is less compared to previous best80.14% but we achieved better results with speed as our GNN model was 1.89xfaster than the previous best avilable method. We also explore multi-level semanticplace classification through GNNs and consider the potential of GraphAttention Networks (GAT) for complex datasets. Here we did a proof of conceptthat multilevel semantic classification is possible with GNNs but it needs moreresearch in this area.
  • Thumbnail Image
    ItemOpen Access
    Neural Network Architectures for Integrated Circuits
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2023) Nagrani, Khyati; Maiti, Tapas Kumar
    This thesis presents the architecture design and implementation of neural networks(NNs) for integrated circuit design. The architecture consists of adders,multipliers, and rectified linear unit (ReLU) blocks. Three architectures, namely,Single-In Single-Out (SISO), Multiple-In Single-Out (MISO), and Multiple-In Multiple-Out (MIMO) are developed. In neural networks, weight values are necessaryand they are supplied from a memory source. The weight values were preparedby training the NNs model on software. Finally, the SISO, MISO, and MIMOneural-networks were taped out. These architectures can be used for intelligentco-processor development.
  • Thumbnail Image
    ItemOpen Access
    VLSI Implementation of Neural Network Driven Augmented FSM
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2022) Patel, Jimmy Kirtikumar; Maiti, Tapas Kumar
    This thesis reports the VLSI implementation of an NN (Neural Network) based emergent behavior model for high-speed robot control. Augmented FSM (Finite- State Machine) is considered to implement the emergent behavior. We performed a system level simulation using our proposed model. For system level simulation, we have used Python base TensorFlow model to implement the Neural Network. Then, we transformed the model to RTL (Register Transfer Level) for circuit simulation. For RTL modeling we have used Verilog (Xilinx, Quartus Prime and iVerilog) and for simulation we have used (Modelsim and GTK wave). In this study, we considered multiple inputs and multiple-outputs NN. Our implementation method improves the speed of execution and accuracy and compares the result with the conventional neural network. For activation function in NN, we implemented sigmoid function with second-order approximation to reduce complexity. We used the walking gesture of the Kondo KHR 3HV robot to verify the model. Finally, we design NN based augmented-AI chip for high-speed robotics applications.
  • Thumbnail Image
    ItemOpen Access
    WSN Network Analysis and Prediction using ML
    (2020) Patel, Charmy Bharatbhai; Sasidhar, P S Kalyan
    Wireless sensor networks are a group of sensors that monitor and record the physical changes of the environment that change rapidly over time. This ability of WSNs help in various fields ranging from the engineering industry to immediate home environments. A sensor node is capable of performing some processing, gathering sensory information, and communicating with other connected nodes in the Network. To make some decisions in this Network, sensors adopt machine learning algorithms. The main aim of this project is to find out the parameters which can increase throughput and decrease the Delay in our Network. Different network sizes have been taken into consideration to find the parameter changes required to meet our above objective. This research project mainly includes the data collection phase observing a network and learning phase. It is simulated for different network scenarios. Three types of machine learning algorithms have been applied: Linear Regression, Neural Network, and Random Forest. By applying these algorithms, we get to know that RandomForest overfits the model, whereas Neural-Network underfits the model because they are non-linear algorithms. Hence we can say that it is showing linear behavior as non-linear algorithms like Neural-Network and RandomForest didn’t help us to estimate the throughput and delay in our Network, and hence they are not suitable.
  • Thumbnail Image
    ItemOpen Access
    Crime information extraction from news articles
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2018) Gohel, Prashant; Jat, P.M.
    In the modern era all news reportings are available in digital form. Most newsagencies put it on their website and are freely available. This motivates us totry extracting some information from online news reporting. While understandingnatural language text for information extraction is a complex task,we hopethat extracting information like crime type, crime location, and some profile informationof accused and victim should be feasible. In this work we pulled about1000 crime news articles from NDTV and Indian Express websites. Hand taggingwas done for crime location and crime types of all articles. Through this workwe show that a combination of LSTM and CNN based solution can be effectivelyused for extracting crime location. Using this technique we get 95.58 % precisionand 94.54 % recall. Further, determination of crime type, we found relatively easier.Through simple key word based classification approach we get 95% precision.We also tried out topic modeling for crime type extraction we do not gain any improvement,and we get 79 % precision. Keywords: crime related named entities,deep learning, neural network, LSTM, CNN, NER, NLP