Abstract

This Full Paper in the Innovative Practice category begins by asking “can algorithms be thought of and taught as dynamical systems?” Our exploration of this idea — Algodynamics — is guided by a vision to achieve convergence between computing and engineering education. The engineering sciences share a common conceptual vocabulary originating in dynamical systems: state spaces, flows, actions, invariants, fixed points, convergence, etc. The goal of algodynamics is to build, ab initio, a framework for understanding and teaching algorithms using concepts from dynamics. This allows us to teach computing and algorithms as an engineering science. Engineers work with models. In algodynamics, models are expressed using transition systems rather than as pseudo- code or programs. This allows a crisp representation of two important classes of computation: sequential algo- rithms as ‘discrete flows’ (iterative systems) and interactive applications as ‘action flows’ (transition systems). The focus of this paper is on the first of these classes, algorithms, but using ideas from the second, viz., tran- sition systems. In this framework, algorithms emerge as convergent iterative systems. Due to their non-interactivity, iterative systems may be hard to understand. The student may trace through the algorithm, but to know how the parts of the algorithm work together requires ‘opening up’ the algorithm and situating it within the more general class of interactive systems. Doing so helps the student to understand the machinery of an algorithm in an incremen- tal and modular way. The student interactively solves the algorithmic problem, experimenting with various strategies along the way. At each stage of the design, interactivity is traded for automation

Abstract

Precise boundary annotations of image regions can be crucial for downstream applications which rely on region-class semantics. Some document collections contain densely laid out, highly irregular and over- lapping multi-class region instances with large range in aspect ratio. Fully automatic boundary estimation approaches tend to be data intensive, cannot handle variable-sized images and produce sub-optimal results for aforementioned images. To address these issues, we propose Bound- aryNet, a novel resizing-free approach for high-precision semi-automatic layout annotation. The variable-sized user selected region of interest is first processed by an attention-guided skip network. The network opti- mization is guided via Fast Marching distance maps to obtain a good quality initial boundary estimate and an associated feature represen- tation. These outputs are processed by a Residual Graph Convolution Network optimized using Hausdorff loss to obtain the final region bound- ary. Results on a challenging image manuscript dataset demonstrate that BoundaryNet outperforms strong baselines and produces high-quality se- mantic region boundaries. Qualitatively, our approach generalizes across multiple document image datasets containing different script systems and layouts, all without additional fine-tuning. We integrate Bound- aryNet into a document annotation system and show that it provides high annotation throughput compared to manual and fully automatic alternatives. Keywords: document layout analysis · interactive · deep learning

Abstract

Handwritten documents are often characterized by dense and uneven layout. Despite advances, standard deep network based approaches for semantic layout segmentation are not robust to complex deformations seen across semantic regions. This phenomenon is especially pronounced for the low-resource Indic palm-leaf manuscript domain. To address the issue, we first introduce Indiscapes2, a new large-scale diverse dataset of Indic manuscripts with semantic layout annotations. Indiscapes2 contains documents from four different historical collections and is 150% larger than its predecessor, Indiscapes. We also propose a novel deep network Palmira for robust, deformation-aware instance segmentation of regions in handwritten manuscripts. We also report Hausdorff distance and its variants as a boundary-aware performance measure. Our experiments demonstrate that Palmira provides robust layouts, outperforms strong baseline approaches and ablative variants. We also include qualitative results on Arabic, South-East Asian and Hebrew historical manuscripts to showcase the generalization capability of Palmira. Keywords: instance segmentation · deformable convolutional network · historical document analysis · document image segmentation · datase

Abstract

We introduce MeronymNet, a novel hierarchical approach for controllable, part-based generation of multi-category objects using a single unified model. We adopt a guided coarse-to-fine strategy involving semantically conditioned generation of bounding box layouts, pixel-level part layouts and ultimately, the object depictions themselves. We use Graph Convolutional Networks, Deep Recurrent Networks along with custom-designed Conditional Variational Autoencoders to enable flexible, diverse and category-aware generation of 2-D objects in a controlled manner. The performance scores for generated objects reflect MeronymNet’s superior performance compared to multiple strong baselines and ablative variants.

Abstract

Control barrier function (CBF) constraints provide a rigorous characterization of the space of control inputs that ensure the satisfaction of state constraints, such as collision avoidance, at all time instants. However, CBFs are highly nonlinear and non-convex and thus, when incorporated within an optimization-based algorithm such as Model Predictive Control (MPC), leads to a computationally challenging problem. Existing works by-pass the computational intractability by collapsing the horizon of the MPC to a single step, although this comes at the cost of severe degradation of performance. In this paper, we present two contributions to ensure the real-time performance of CBFs based MPC over long horizons in the context of multi-robot collision avoidance. First, we propose a customized Project Gradient Descent Method that incurs minimal computational overhead over existing one-step approaches but leads to a substantial improvement in trajectory smoothness, time to reach the goal, etc. Our second contribution lies in applying the proposed MPC to both quadrotors and fixed-wing aerial aircrafts (FWA). In particular, we show that the formulation for the quadrotors can be readily extended to the latter by deriving additional CBFs for the curvature and forward velocity constraints. We validate our algorithm with an extensive simulation of up to 10 robots in challenging benchmark scenarios.

Abstract

The intensity of online abuse has increased in recent years. Automated tools are being developed to prevent the use of hate speech and offensive content. Most of the technologies use natural language and machine learning tools to identify offensive text. In a multilingual society, where code-mixing is a norm, the hate content would be delivered in a code-mixed form in social media, which makes offensive content identification, further challenging. In this work, we participated in the EACL task to detect offensive content in the code-mixed social media scenario. The methodology uses a transformer model with transliteration and class balancing loss for offensive content identification. In this task, our model has been ranked 2 nd in Malayalam-English and 4 th in Tamil-English code-mixed languages.

Abstract

Code-mixing (CM) is a frequently observed phenomenon that uses multiple languages in an utterance or sentence. There are no strict grammatical constraints observed in codemixing, and it consists of non-standard variations of spelling. The linguistic complexity resulting from the above factors made the computational analysis of the code-mixed language a challenging task. Language identification (LI) and part of speech (POS) tagging are the fundamental steps that help analyze the structure of the code-mixed text. Often, the LI and POS tagging tasks are interdependent in the code-mixing scenario. We project the problem of dealing with multilingualism and grammatical structure while analyzing the code-mixed sentence as a joint learning task. In this paper, we jointly train and optimize language detection and part of speech tagging models in the code-mixed scenario. We used a Transformer with convolutional neural network architecture. We train a joint learning method by combining POS tagging and LI models on code-mixed social media text obtained from the ICON shared task.

Abstract

This paper presents an ultra-low power CMOS voltage reference (CVR) which is free of amplifier and resistor. A current generator circuit is used to generate a supply independent current to bias the active load in the temperature compensation circuit. Drain-source voltage of two critical MOSFETs is made equal in the current generator circuit by using a feedback arrangement to ensure a high PSRR of -75dB for the reference voltage. Temperature compensation is achieved by using the complementary to absolute temperature (CTAT) nature of gatesource of a MOSFET operating in the subthreshold region and proportional to absolute temperature (PTAT) nature of conventional composite pair architecture. The proposed CVR is designed in TSMC 180nm technology. The circuit works desirably for a supply range of 0.85V to 2.3V while generating a reference voltage of around 0.68V. Maximum temperature coefficient (TC) of 184 ppm/˚C and minimum TC of 69 ppm/˚C are noted for mismatch and process variations in the Monte-Carlo simulation for 1000 samples. The circuit consumes only 46nW of power, which makes it suitable for ultra-low power applications. Keywords— CMOS voltage reference, Line sensitivity, Subthreshold region, Temperature compensation.

Abstract

In this paper, we study FPGA based pipelined and superscalar design of two variants of conjugate gradient methods for solving Laplacian equation on a discrete grid; the first version corresponds to the original conjugate gradient algorithm, and the second version corresponds to a slightly modified version of the same. In conjugate gradient method to solve partial differential equations, matrix vector operations are required in each iteration; these operations can be implemented as 5 point stencil operations on the grid without explicitely constructing the matrix. We show that a pipelined and superscalar design using high level synthesis written in C language leads to a significant reduction in latencies for both methods. When comparing these two, we show that the later has roughly two times lower latency than the former given the same degree of superscalarity. These reductions in latencies for the newer variant of CG is due to parallel implementations of stencil operation on subdomains of the grid, and dut to overlap of these stencil operations with dot product operations. In a superscalar design, domain needs to be partitioned, and boundary data needs to be copied, which requires padding. In 1D partition, the padding latency increases as the number of partitions increase. For a streaming data flow model, we propose a novel traversal of the grid for 2D domain decomposition that leads to 2 times reduction in latency cost involved with padding compared to 1D partitions. Our implementation is roughly 10 times faster than software implementation for linear system of dimension 10000 × 10000. Index Terms—FPGA, High Level Synthesis, Conjugate Gradient, Laplace System, Pipelining, Superscalarity.

Abstract

As a big data application, extreme multilabel classification has emerged as an important research topic with applications in ranking and recommendation of products and items. A scalable hybrid distributed and shared memory implementation of extreme classification for large scale ranking and recommendation is proposed. In particular, the implementation is a mix of message passing using MPI across nodes and using multithreading on the nodes using OpenMP. The expression for communication latency and communication volume is derived. Parallelism using work-span model is derived for shared memory architecture. This throws light on the expected scalability of similar extreme classification methods. Experiments show that the implementation is relatively faster to train and test on some large datasets. In some cases, model size is relatively small. Code: https://github.com/misterpawan/DXML