<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Jaisimha Manipatruni on Medium]]></title>
        <description><![CDATA[Stories by Jaisimha Manipatruni on Medium]]></description>
        <link>https://medium.com/@jaisimha34?source=rss-26a6ca2ea3e4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*P9dM700Kfgxs7vmXdMWYjA@2x.jpeg</url>
            <title>Stories by Jaisimha Manipatruni on Medium</title>
            <link>https://medium.com/@jaisimha34?source=rss-26a6ca2ea3e4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 16:45:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@jaisimha34/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Deep Learning-based Arrhythmia Detection -A Deep-Belief Network Approach]]></title>
            <link>https://medium.com/@jaisimha34/deep-learning-based-arrhythmia-detection-a-deep-belief-network-approach-b0847ab66641?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0847ab66641</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[health]]></category>
            <category><![CDATA[science]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Thu, 16 Jan 2020 18:55:34 GMT</pubDate>
            <atom:updated>2020-01-16T18:55:34.835Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>According to the World Health Organization surveys, heart diseases are one of the most important reasons which cause death. Heart disease symptoms depend on what type of heart disease you have. An electrocardiogram (ECG) is a nonlinear and non-stationary diagnostic signal that is important for cardiac disorders.</p><p><strong><em>Arrhythmia</em></strong><em> </em>is a problem concerning the abnormal rhythm and rate of heartbeats. The heart can beat too fast, too slowly, or inconsistently in different types of arrhythmias, which may feel like antagonism affection or fluttering. Several types of arrhythmia are harmless, but some of them refer the cardiac disorders that may cause death. The ECG is a popular diagnosis tool which is of the primary importance for cardiologists.</p><h4><strong><em>Deep Learning Approach</em></strong></h4><p>Deep learning (DL) is an effective and high-performance machine learning algorithm which is gaining popularity. Frequently used analyses of the DL are used in <strong><em>image processing, speech and natural language processing</em></strong> processes. In this analysis, <strong><em>Deep Belief Networks (DBN)</em></strong>, which is an adaptable DL algorithm, is utilized to classify the heartbeats<em> </em>from different classes of arrhythmia using ECG waveform as the input of the structure.</p><h4><strong>The Data Set Used</strong></h4><p>In this analysis, the <strong><em>MIT-BIH arrhythmia database (MADB)</em></strong> is utilized.This database has been used for evaluating arrhythmia detection and classifying the arrhythmia types. MADB contains 48 long-term ECGs from 25 men aged 32–89 years, and 22 women aged 23–89 years; each has 11-bit resolution with 360 Hz sampling frequency. The heartbeats are labelled as five main arrhythmia types defined by the Association for the Advancement of Medical Instruments (AAMI) standard</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/504/1*ocNRuQXKITPROdl_So-Wcw.jpeg" /></figure><p>AAMI classifies heartbeats into normal beats (N), supraventricular ectopic heartbeats (S), ventricular ectopic heartbeats (V), fusion heartbeats (F), and unknown heartbeats (Q).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/813/1*6-Xmn1bWUd8pJj1PV7TGoA.jpeg" /></figure><h4><strong>Deep Belief Networks</strong></h4><p>This approach introduces deep learning (DL) application for <strong>automatic arrhythmia classification</strong>. The proposed model consists of a multi-stage classification system of raw ECG using DL algorithms. The <strong>DBN </strong>is one of the most effective DL algorithms which has a <strong><em>greedy layer-wise training phase</em></strong> [. The DBN is composed of both <strong><em>Restricted Boltzmann Machines (RBM)</em></strong> or an <strong><em>autoencoder based layer-by-layer unsupervised pre-training procedure </em></strong>and neural network-based supervised training. Considering RBM with input layer activations ‘v’(for visible units) and hidden layer activations ℎ (hidden units), the bias of the visible unit unit ‘b’ ,bias of hidden unit ‘c’.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/282/1*P7w9LqZXeJ9kXctjwaz7Mg.jpeg" /><figcaption>Representing Equation</figcaption></figure><h4><strong>Results and Analysis</strong></h4><p>The training set of the DBN-based automatic arrhythmia classification model includes 4,077 of ECG waveforms from various types of heartbeat classes distributed homogeneously. The DBN-based multistage model is tested using 2,000 of ECG waveforms.</p><h4>The ECG Wave Form With Faulty Datapoints</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/667/1*hnk0yysxZg43lsBprpTm7A.jpeg" /><figcaption>Faulty Data Point ECG Wave</figcaption></figure><p>The proposed multistage DBN model separates N, S, V, F, and Q types of arrhythmias, respectively. Four of the DBN models are used in the proposed system. The RBM based greedy layer-wise pre-training is used in this model at the unsupervised learning stages of all DBNs with <strong><em>5 epochs</em></strong>. The parameters of the RBM were denoted by iterations. The models were tested with a limited number of the parameters and the highest classification performances are given. The learning rate of the model is 3 and the softmax output function was utilized constantly. The proposed multistage arrhythmia classification model consists of <strong><em>4 DBN structures </em></strong>with various numbers of hidden units. The <strong><em>DBN1 has 2 hidden layers with 100–260 hidden units</em></strong>; the<strong><em> DBN2 has 3 hidden layers with 230–520–210 hidden units; the DBN3 has 2 hidden layers with 120–240 hidden units; and the DBN4 has 2 hidden layers with 70–190 hidden units. </em></strong>The four DBN structures are connected sequentially and have the ability to separate five classes of arrhythmia types defined by ANSI/AAMI.</p><h4>Conclusion and AI model Accuracy</h4><p>The waves plot a regular form in normal sinus rhythm. Any obvious changes occurring in indicates the irregularity or arrhythmia in heartbeats.</p><p>The meaningful data points for arrhythmia classification are thickened between S-T waves and P-Q waves for the proposed DBN-based multistage classification model</p><p>The proposed DBN-based Arrhythmia classification has discriminated five types of heartbeats with a <strong><em>very highest accuracy of 95.05%. </em></strong>This achievement proves the success and efficiency of the Deep Belief Network algorithm in raw ECG signals for Cardiac Health Conditions.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0847ab66641" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Performance Benchmarking Quantum Hardware for Deep Learning Boltzmann Models]]></title>
            <link>https://medium.com/@jaisimha34/performance-benchmarking-quantum-hardware-for-deep-learning-boltzmann-models-c671ba8f702f?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/c671ba8f702f</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Wed, 02 Oct 2019 14:45:59 GMT</pubDate>
            <atom:updated>2019-10-02T14:54:48.043Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction :</h3><p>Quantum annealing (QA) is a hardware-based heuristic optimization and sampling method applicable to discrete undirected graphical models. While similar to simulated annealing, QA relies on quantum, rather than thermal, effects to explore complex search spaces. Here we report on the ability of recent QA hardware to accelerate training of fully visible Boltzmann machines.</p><p><strong>Quantum Annealing:</strong> QA uses quantum-mechanical processes to minimize and sample from energy-based models. The machine implements the Ising model energy function</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/754/1*9OSS4eUtQ58gndULvqfltQ.jpeg" /><figcaption>Ising Model Energy Function</figcaption></figure><p>with variable connectivity defined by a graph G = (V, E). The 2000-qubit D-Wave system allows for up to |V| = 1152 variables with sparse bipartite connectivity</p><h3><strong>Chimera Graphs :</strong></h3><p>The connectivity graph of the D-Wave device is called Chimera and denoted Cn and it is visualised as below</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/537/1*vs3OEG7DMvl867-WDxheXg.jpeg" /><figcaption>Graph Unit</figcaption></figure><p>Cn consists of an n × n array of K4,4 unit cells with a connection between unit cells. he tree-width of C12 graph is 48, so the exact inference is practically impossible. It is simple to convert ±1 valued spins sv to Boolean-valued variables xv = (1 + sv)/2 so that E(s) also defines a BM with energy E(x) and the same sparse bipartite connectivity.</p><h3><strong>The Ising Hamiltonian Parameters</strong></h3><p><strong><em>Finite temperature</em></strong><em>:</em> QA hardware does not operate at zero temperature. In units where the parameters lie in the interval −1 ≤ hv ≤ 1 and −1 ≤ Jv,v0 ≤ 1, the effective hardware temperature THW is problem dependent and usually between 1/5 and 1.</p><p><strong><em>Dynamics:</em></strong> The quantum mechanical evolution of the annealing process cannot be simulated at large scales (even for idealized models), and quantum effects can cause significant deviations from the classical Boltzmann distribution.</p><h3>Performance Graphs for Local Minima search in QA :</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/646/1*ussD5hkOanS0u7iHYhB2Vw.jpeg" /></figure><p>Probabilities of local optima for 10^5 QA and MCMC annealing samples. Red bars represent exact Boltzmann probabilities of local optima, while blue and green bars represent empirical probabilities of QA and MCMC samples.</p><p><strong><em>Methodology Used :</em></strong></p><p>Quantum Simulation for training the model has been done using modular graph training on Distributed Deep Learning Style.</p><p>Blocked Gibbs sampling without annealing performed much more poorly. The used 100 random spin-reversal transformations as suggested by D-Wave to mitigate parameter migrated to minima.</p><h3>Conclusion :</h3><p>The computational benefits of QA are demonstrated in artificial problems constructed to have high-energy barriers between modes, but which were small enough to yield exact results.</p><p>The Energy Based RBM Deep Learning model result distribution has been displayed in this article but the Model Architecture and Code were not revealed by me deliberately.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c671ba8f702f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Embedding Photonic Chip on Scalable GPU Multicore Architecture]]></title>
            <link>https://medium.com/@jaisimha34/embedding-photonic-chip-on-scalable-gpu-multicore-architecture-6e38acd05294?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/6e38acd05294</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[gpu]]></category>
            <category><![CDATA[photonics]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Mon, 12 Aug 2019 13:02:42 GMT</pubDate>
            <atom:updated>2019-08-12T13:02:42.433Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/675/1*LKlGnmXl9V0sReoHBE2UrA.jpeg" /><figcaption>Physical Layout of Embedded Photonic GPU</figcaption></figure><p>A little over a decade ago, GPUs were fixed-function processors built around a pipeline that was dedicated to rendering 3-D graphics. In the past decade, as the potential for GPUs to provide massive compute parallelism became apparent, the software community developed programming environments to leverage these massively-parallel architectures.NVIDIA and AMD, two of the leading graphics vendors today, tailor their GPU designs for general-purpose high-performance computing by providing higher compute throughput and memory bandwidth.</p><p>Since GPU applications generally exhibit little L1 temporal locality, the communication between the L1 and L2 caches becomes the main source of traffic on the GPU’s on-chip interconnection network.</p><p>As the number of CUs increases in each future generation of GPU systems, latency in the on-chip interconnection network becomes a major performance bottleneck on the GPU.</p><h3>Silicon- Photonic Network on-chip for solving latency</h3><p>Silicon-photonic link technology has been proposed as a potential replacement for an electrical link technology for the design of NoC architectures for future manycore systems Silicon-photonic links have the potential to provide significantly lower latency and an order of magnitude higher bandwidth density for global communication in comparison to electrical links. We take advantage of photonic link technology in designing an NoC for GPU systems.</p><h3>Hardware Design and Embeddings on Multicores</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/851/1*HLLxSSJdFz2P3YS-QikcBA.jpeg" /><figcaption>Photonic Link Components — Two point-to-point photonic links implementation</figcaption></figure><p>This is a generic silicon-photonic channel with two links multiplexed onto the same waveguide. These two silicon-photonic links are powered using a laser source. The output of the laser source is coupled into the planar waveguides using vertical grating couplers. Each Electrical-to-Optical (E/O) and Optical-to-Electrical (O/E) conversion stages introduce a 1-cycle latency. The time-of-flight for the data on the photonic link is 1 cycle. The serialization latency, on the other hand, depends on the bandwidth of the photonic link. Contention and queuing delays are other contributing factors to the overall latency of transmissions. We consider all these forms of latency in our simulation framework.</p><p>We evaluate the scalability of a silicon-photonic NoC on a larger GPU system with 128 CUs, comparing our proposed hybrid photonic crossbar against a competitive electrical 2D-mesh NoC.</p><h3>Target 128 CU GPU System and Performance Enhancement</h3><p>We consider a forward-looking GPU design that uses the 32-CU Radeon HD 7970 GPU as a building block and increases the number of CUs. In this new architecture, we quadruple the number of CUs to 128 while keeping the same CU design as in the 32-CU GPU for individual CUs. The 128-CU GPU also quadruples the number of memory components in the HD 7970 GPU chip: 128 L1 vector caches and can observe a high boost in performance</p><h3>Performance Benchmarks</h3><p>Although not every detail of the interconnect used in the AMD Radeon HD 7970 chip has been publicly disclosed by me here are the performance benchmarks from embedded simulation workloads after Photonic Embedding</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/848/1*3cj1L4f-27TvRytKV1HlKA.jpeg" /><figcaption>Potential improvements for various workloads running on a 128-CU GPU</figcaption></figure><h3>Conclusion</h3><p>The simulation-based analysis shows that applications that are bandwidth sensitive can take advantage of a photonic hybrid NoC to achieve better performance while achieving an energy-delay value that is lower than the traditional electrical NoC. In the AMD Southern Islands GPU chip with 128 CUs, the proposed photonic hybrid NoC increases application performance by up to 6× (2.7× on average) while reducing ED2P by up to 99% .</p><p>My future work includes moving forward, I plan to explore techniques for run-time power management of photonic NoCs to extend this improvement to all applications.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6e38acd05294" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Quantum Computer Architecture Based on Optically Controlled Spin Mechanics]]></title>
            <link>https://medium.com/@jaisimha34/quantum-computer-architecture-based-on-optically-controlled-spin-mechanics-d66bc08ad7c7?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/d66bc08ad7c7</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Fri, 12 Jul 2019 12:48:50 GMT</pubDate>
            <atom:updated>2019-07-12T12:48:50.575Z</atom:updated>
            <content:encoded><![CDATA[<h3>Quantum Computer Architecture Based on Optically Controlled Spins</h3><h3>INTRODUCTION</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/749/1*tfla79h2keJr34_V4Na7gQ.jpeg" /><figcaption>Layered Stack Architecture Over View</figcaption></figure><p>Quantum computing as an engineering discipline is still in its infancy. Although the physics is well understood, developing devices which compute with quantum mechanics is technologically daunting. While experiments to date manipulate only a handful of quantum bits we consider what effort is required to build a large-scale quantum computer. This objective demands more than a cursory estimate of the number of qubits and gates required for a given algorithm. One must consider the faulty quantum hardware, with errors caused by both the environment and deliberate control operations; when error correction is invoked, classical processing is required; constructing arbitrary gate sequences from a limited fault-tolerant set requires special treatment, and so on.</p><p>An architecture decomposes complex system behaviours into a manageable set of operations. A layered architecture does this through layers of abstraction where each embodies a critical set of related functions. For our purposes, each ascending layer brings the system closer to an ideal quantum computing environment</p><h4>Layered Control Stack Architecture</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/693/1*SOfKaXO6eqQBo1yiVXSNcg.jpeg" /><figcaption>Layered Stack Architecture</figcaption></figure><p>Layered architectures are a conventional approach to solving such engineering problems in many fields of information technology, and presents a layered architecture for quantum computer design software. This architecture, which describes the physical design of the quantum computer, consists of five layers, where each layer has a prescribed set of duties to accomplish. The interface between two layers is defined by the services a lower layer provides to the one above it. To execute an operation, a layer must issue commands to the layer below and process the results.</p><h4>Primary Control Cycle of Quantum Architecture</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/1*i11DpqarkWoyYCDP8Y7dVQ.jpeg" /><figcaption>Pipeline Control Cycle</figcaption></figure><p>The primary control cycle defines the dynamic behaviour of the quantum computer in this architecture since all operations must interact with this loop. The principal purpose of the control cycle is to successfully implement quantum error correction. The quantum computer must operate fast enough to correct errors; still, some control operations necessarily incur delays, so this cycle does not simply issue a single command and wait for the result before proceeding — pipelining is essential. Layers 1 to 4 interacts in the loop, whereas the Application layer interfaces only with the Logical layer since it is agnostic about the underlying design of the quantum computer.</p><h4>Optically Controller Spin and The QuDOS hardware platform</h4><p>The layered framework for quantum computing was developed in tandem with a specific hardware platform, known as QuDOS (quantum dots with optically controlled spins). The QuDOS platform uses electron spins within quantum dots for qubits. The quantum dots are arranged in a two-dimensional array</p><p>The QuDOS design is a promising candidate for large-scale quantum computing, beginning with an analysis of the hardware</p><h4>Hardware performance summary</h4><p>These are the quantum processes which are the building blocks of quantum information operations in Layers 2 and above. For a complete quantum processor, however, one would also have to consider the classical control hardware and the engineering concerns, such as delays, which may occur in a large system</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1008/1*v2nWnNrGdlDmUzV6QpaUaw.jpeg" /><figcaption>Performance Table</figcaption></figure><h4>Conclusion</h4><p>Designing a quantum computer requires viewing the system as a whole, such that tradeoffs and compatibility between component choices must be addressed. A holistic picture is equally important for comparing different quantum computing technologies, such as ion traps or superconducting circuits. This work illustrates how to approach the complete challenge of designing a quantum computer so that one can adapt these techniques to develop architectures</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d66bc08ad7c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deep Learning Compiler Architecture]]></title>
            <link>https://medium.com/@jaisimha34/deep-learning-compiler-architecture-e9e3e5ce99cb?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9e3e5ce99cb</guid>
            <category><![CDATA[hardware]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Mon, 27 May 2019 13:00:41 GMT</pubDate>
            <atom:updated>2019-05-27T13:00:41.011Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>General Architecture of Deep Learning Customised Compiler Stack</blockquote><p>Deep Learning Virtual Machine (DLVM) is a compiler infrastructure designed for modern deep learning systems. DLVM is designed to apply a multi-stage compiler optimization strategy to both high-level linear algebra and low-level parallelism, perform domain-specific transformations and relieve the overhead from front-end languages, and serve as the host for research and development of DSLs for neural networks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/570/1*715VoaRLat5sHBqGE9BGOA.jpeg" /><figcaption>Software stack of the DLVM infrastructure. Blue components are the compiler framework.</figcaption></figure><h4>DLVM CORE</h4><p>The DLVM consists of a virtual instruction set, control flow graph and data flow representation. Passes are functions that traverse the intermediate representation of a program, either producing useful results as analyses of the program (analysis passes) or mutating the program for differentiation and optimizations (transform passes).</p><p>The DLVM has a high-level type system with tensor as a first-class type. The DLVM virtual instruction set includes domain-specific primitive math operators, as well as general-purpose instructions for memory management, control flow and function application. Domain-specific instructions include element-wise unary operators, such as tanh and negate element-wise binary operators, such as add and power, and complex operators such as dot, transpose, and convolve. All element-wise binary operators support broadcasting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/556/1*gZwgNjWi_qJof6jmxhZ3MQ.jpeg" /><figcaption>Example code in DLVM intermediate representation.</figcaption></figure><h3>Conclusion</h3><p>While well intentioned, the remaining vast majority of other frameworks have failed to observe that the problem of algorithmic differentiation and converting a neural network into efficient executable code is, at its core, a compilers problem. As a result, important issues of extensibility and optimization have been addressed in less than optimal fashion in such frameworks. Nevertheless, several such frameworks have achieved wide adoption. The principled application of optimizing compiler techniques will lead to substantial improvements in the tools available to deep learning researchers. DLVM and its associated front-end DSLs have a major role to play in this future. Our existing implementation utilizes LLVM to target NVIDIA GPUs. In our ongoing work, we plan to substantially increase the number of supported hardware architectures by utilizing HPVM as an additional back-end, while simultaneously supporting the use of DLVM by existing front-end DSLs such as TensorFlow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e9e3e5ce99cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Chartered Accountants need to embrace Artificial Intelligence]]></title>
            <link>https://medium.com/@jaisimha34/why-chartered-accountants-need-to-embrace-artificial-intelligence-109cefe40574?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/109cefe40574</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[careers]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Thu, 20 Dec 2018 10:50:19 GMT</pubDate>
            <atom:updated>2018-12-20T18:14:25.040Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/339/1*PWDYdG8rO4pJpw-KWfsSrg@2x.jpeg" /></figure><h4>First let’s look at the Spectrum of Ai</h4><blockquote>The Spectrum of Artificial Intelligence</blockquote><p>Part of the challenge with an emerging technology is that there is often an unclear definition of what technology is, and what it is not. As vendors’ marketing departments seek to leverage the hype to drive sales, they often start referring to technology buzzwords in a loose sense and cause additional blurring of the definition.</p><p>To help properly set the stage for analyzing the implications of AI on accountants, I’m going to use the broad AI technology categories of:</p><p>Machine learning: the ability of the computer to recognize and apply patterns, derive its own algorithms based on those patterns, and refine those algorithms based on feedback.</p><p>Deep learning: the ability of the computer to identify relationships and associations, and apply those in similar circumstances (this partly what our brain does).</p><p>Machine reasoning: the ability of the computer to apply its “understanding” of data, relationships, rules, etc., to “think” though the implications of a particular set of information and provide some analysis or interpretation.</p><p>Natural language processing: the ability of the computer to “understand” human speech.</p><p>Computer vision: the ability of the computer to “see” images and “recognize” people, things, activities, and states (e.g. happy, sad, in motion, etc.) in those images.</p><p>Of these categories, machine learning has the broadest available applications and its functionality can most greatly supplement the faculties of an accountant, so this article will focus on machine learning.</p><h4>You’ve Already Experienced Machine Learning</h4><p>Machine learning is good at “inductive reasoning” – where based on a set of existing data points or examples, a computer can figure out what the “rules” are to determine a result. Take a step back to your statistics class in university and you may remember techniques like linear regression, the measurement of the correlation, and reliability of various data points. At a very basic level, these are the types of analyses that machine learning algorithms are applying to predict outcomes. With the computational power of a computer behind it, machine learning can process thousands of data points about a given set of situations to figure out which ones are relevant, and which are not, and then apply the inferred rules to another similar set of data to predict outcomes.</p><p>Amazon’s, Kindle’s, and Netflix’s suggestions to users are a great example of this. Netflix is able to use your ratings of other shows in its library and data points, like genre, director, actors, etc., to predict whether you will like another show. Kindle does something similar for books. Amazon uses product views, other shoppers’ purchase history, and complimentary items to those in your cart to suggest additional products you may like. Whether you realize it or not, you’ve already had machine learning applied to try and predict what you may like.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/358/1*QoT7VVdZcHHAgZmnVtv2bg@2x.jpeg" /></figure><h3>Machine Learning Implications for Auditors</h3><p>There is a high potential for machine learning to provide augmented analyses to auditors. Note that I did not say that it would replace auditors – machine learning is just another tool in the auditor’s belt of Computer Assisted Auditing Tools and Techniques (CAATTs).</p><p>Instead of sampling data, auditors can push an entity’s entire ledger through automated analysis. This, by the way, is not AI or machine learning; this is a capability that already exists in tools like IDEA and ACL. These tools can perform a variety of analyses, designed by humans, and then provide lists of exceptions for the auditor to evaluate. Machine learning comes into play as the auditor confirms the exception or invalidates that exception and the machine learns to “look” at the auditor’s conclusions and try to identify additional data points about the positives or negatives to apply to additional exceptions it identifies. In this way it learns to better identify exceptions.</p><p>The data bias risk in this application is that if an auditor incorrectly clears items that should be confirmed as exceptions, machine learning would start to clear other items that should be exceptions. So a review process must be put in place to ensure that cleared exceptions really are not exceptions. The converse is also true for confirmed exceptions that should not be exceptions.</p><p>In a more advanced application, a set of transactions could be provided to an AI tool and machine learning would identify the patters in the transactions and be able to identify what “normal transactions” look like. Using this method, it would then identify those exceptions that don’t match the norm as exceptions. This application of machine learning is also subject to data bias since its picture of a normal transaction is based on the set of data provided. If the data set that was used also happened to have a high incidence of fraudulent transaction, then those fraudulent transactions may be interpreted to be normal transactions since they are highly present in the data set that the AI learns from.</p><p>There is definitely a future need for a human auditor even as machine learning starts to augment audit procedures. The auditor role, for both internal and external auditors, will switch from performance of the procedures to design of the procedures, interpretation of the results, and monitoring the effectiveness of the interpretation.</p><h3>Machine Learning Implications for Management Accountants</h3><p>The implications of machine learning for management accountants and other professional accountants working in business and government is even greater than it is for auditors. In addition to machine learning being applied within finance, it may also be applied in other parts of the organization and management accountants must ensure that there are proper governance and internal controls applied to machine learning throughout the organization.</p><p>Within the controllership function, machine learning may be applied to help with the classification of transactions. Inductive reasoning could be applied to the source data of historical transactions to help “predict” the classification of additional transactions as they are recorded. Since the products of many vendors have a fairly consistent natural classification, for the most part this is ok. However, there are some vendors that could be placed in different natural classifications depending on how their product is used. Take, for example, an email newsletter tool used to provide marketing and promotional emails to customers and classified as advertising spend. However, if it were used to generate employee newsletters, it may instead be classified as employee relations or an IT expense. Human validation of the classification for vendors with this risk may be necessary depending on the materiality of the potential error.</p><p>When used as part of financial planning &amp; analysis (FP&amp;A), machine learning can be used to analyze data to define or refine data models used for forecasting. The quality of the data set being used and the risk of inherent biases may again impact the quality of the predictions provided by machine learning. FP&amp;A accountants must exercise care due to the impacts of the data sets used for their models.</p><h3>Opportunities in Enterprise Use of Machine Learning</h3><p>Organizations that are exploring machine learning must also address the additional governance and internal controls considerations for the associated risks. As departments outside of finance seek to employ machine learning, FP&amp;A accountants have a big opportunity to provide their data analysis and modeling expertise to help other departments develop their applications of machine learning.</p><p>This is not an area that can be addressed by IT alone. A holistic view of the data, processes, and use of information provided by machine learning must be obtained. For each project, accountants in finance and internal audit must be sure to understand the compliance requirements, and assess the design of controls to mitigate machine learning risks from biased data.</p><p>Internal auditors also play an important on-going role in evaluating the design and effectiveness of the governance and internal controls over machine learning, and in evaluating the effectiveness of the methods chosen to reduce the risk of negative impacts from biased data.</p><h3>Accountants Must Embrace Machine Learning</h3><p>AI as an accountant replacement myth is really just more of the hype that will be proven wrong. Instead, increased use of AI will allow accountants to focus on providing better decision support rather than on data gathering and manual analyses. Increased use of AI will also require accountants to step up and address associated risks with AI through effective governance and internal controls.</p><p>Accountants need to look at both how we can leverage machine learning to facilitate our role as auditors and accountants. There is also a large opportunity beyond the finance context to guide other departments in their use of machine learning and help with the design of internal controls over their applications.</p><p>The corporate governance function must also be adjusted to address business strategy alignment. of the risks presented by machine learning technology. A sub-function focused on data governance should be established to address both data bias risks as well as compliance risks, like privacy. There are areas that IT cannot address alone since they do not have the risk and controls expertise that accountants do.</p><p>By embracing machine learning as a tool, accountants can shift where we’re spending our time from performing menial data preparation and analyses to the drawing of insights from those analyses. Accountants’ expertise in controls design and understanding data biases can also be used to serve other departments in the organization as the departments seek to embrace machine learning.</p><p>Machine learning provides an unprecedented opportunity for accountants and we must embrace it to enhance both our careers and the competitive advantage it can provide to the organizations that we serve.</p><blockquote>HOPE YOU LIKED THIS</blockquote><p>Thank you .</p><blockquote>Lot more coming on Artificial Intelligence</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=109cefe40574" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Psychology and Artificial Intelligence]]></title>
            <link>https://medium.com/@jaisimha34/psychology-und-artificial-intelligence-c08cb72b0918?source=rss-26a6ca2ea3e4------2</link>
            <guid isPermaLink="false">https://medium.com/p/c08cb72b0918</guid>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <dc:creator><![CDATA[Jaisimha Manipatruni]]></dc:creator>
            <pubDate>Wed, 12 Dec 2018 17:09:46 GMT</pubDate>
            <atom:updated>2018-12-13T18:08:11.136Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*Bc3ugw_JM6KSa5F1_MkC-A@2x.jpeg" /></figure><p>Psychology is one of the basic sciences of artificial intelligence (AI). The founder of the psychology is Wilhelm Wundt (1832–1920), who engaged in empirical ways, and was interested in the thinking processes during his work. The main goal of the science is to understand human thinking (in AI the Human Computing deals with this )but also an other question rose: How can we solve psychological, human-near questions or problems with artificial intelligence methods? If we possess the detailed description of the brain, it will enable us to model this on computer, for example: a broader understanding of natural language processing, understanding emotional intelligence, sight.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/306/1*jrSkUnQOk4Xz8InHAdTDvg@2x.jpeg" /></figure><h4>The Cognitive Systems Movement!</h4><p>However, the original problems remain and some researchers are committed to pursuing them.</p><p>Because “AI” now has such limited connotations, we will refer</p><p>to cognitive systems as the paradigm that:</p><ul><li>Designs, constructs, and studies computational artifacts that</li><li>exhibit human-like intelligence.</li><li>Brachman and Lemnios (2002) promoted this term for their</li><li>DARPA funding initiative in the area.</li><li>We can distinguish the cognitive systems movement from most current AI work by five characteristics.</li></ul><h4>AI and Cognitive Psychology</h4><p>In contrast, much early AI research was inspired by, and gained insights from, studies of human thinking.</p><p>This link has produced many of the most powerful ideas about the computational character of the mind:</p><ul><li>Symbol structures and processing</li><li>Heuristic search in problem solving</li><li>Knowledge, expertise, and rule-based systems</li><li>Unified cognitive architectures</li><li>Artificial intelligence can grow even stronger by drawing on its deep psychological roots.</li></ul><p>Hope you liked it :)</p><p>Lot more coming</p><p>Thank you !!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c08cb72b0918" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>