<?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 Paul Woody on Medium]]></title>
        <description><![CDATA[Stories by Paul Woody on Medium]]></description>
        <link>https://medium.com/@paul.woody?source=rss-fac77cf82aef------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*_BT42JFRDxmL3PSv0Gpbuw.jpeg</url>
            <title>Stories by Paul Woody on Medium</title>
            <link>https://medium.com/@paul.woody?source=rss-fac77cf82aef------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 21:14:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@paul.woody/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[Using Convolutional Neural Networks to Classify Street Signs]]></title>
            <link>https://medium.com/analytics-vidhya/using-convolutional-neural-networks-to-classify-street-signs-eef7bd6a200a?source=rss-fac77cf82aef------2</link>
            <guid isPermaLink="false">https://medium.com/p/eef7bd6a200a</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <dc:creator><![CDATA[Paul Woody]]></dc:creator>
            <pubDate>Mon, 25 Nov 2019 21:24:06 GMT</pubDate>
            <atom:updated>2019-11-27T12:24:36.515Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3bnhGykmfZPbwl5AZ2PLYQ.jpeg" /><figcaption>Image Source: yonah.org</figcaption></figure><p>Since the invention of the automobile, manufacturers have steadily added more safety features and improved car design over time with the goal of keeping drivers safer on the road. Automotive manufacturers have spent millions of dollars researching safety improvements for seatbelts, tires, and pretty much every car piece or part imaginable. Despite all of this investment, driving remains substantially more fatal than alternatives such as <a href="https://www.theguardian.com/world/2019/jan/02/plane-crash-deaths-jump-sharply-in-2018-but-fatalities-still-rare">air travel</a> in 2019. According to the <a href="https://www.nsc.org/road-safety/safety-topics/fatality-estimates">National Safety Council</a>, approximately 40,000 people died in automotive accidents in the United States alone in 2018. In fact, there were a total of ~500 deaths resulting from plane crashes recorded globally in 2018 — that’s <strong>80 times</strong> <strong>fewer deaths</strong> when compared to car crash fatalities in the US only.</p><p>There are many factors that likely contribute to this drastic difference, such as the difficulty in training and certification for commercial pilots, compared to motor vehicle drivers and the amount of cars on the road vs. planes. Despite this, the difference in mortality remains, in my opinion, pretty staggering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*OE7MmQO3O0Pb9E6a.jpeg" /></figure><blockquote>Today, companies such as Tesla, Volvo, and many others are investing into technology such as Artificial Intelligence that seeks to correct the root cause of the issue, and the one we’ve previously ignored — <strong>human performance behind the wheel.</strong></blockquote><p>Perhaps the most exciting advancement in automotive technology since the invention of the car itself is the birth of self-driving, or autonomous vehicles. Once the topic of Science Fiction, self-driving cars have been made possible through the power of Convolutional Neural Networks (CNNs) and other modeling techniques. But, before we hand over the keys entirely, there are 3 major challenges we must overcome.</p><p>We must:</p><p>&gt; 1. Obtain the data (sensors, cameras)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/1*juuc_tJPtezMivSYcqH-gg.jpeg" /><figcaption>Image source: <a href="https://www.sae.org/*">www.sae.org</a></figcaption></figure><p>&gt; 2. Process the data (AI — we’ll focus on this part)</p><p>&gt; 3. Act on the data (drive the car)</p><p>To highlight a very small sliver of the technological challenges faced by companies like Tesla in bringing autonomous cars to the public, we’ll focus exclusively on a simplified application of step 2, <strong>processing image data in order to determine what it is we’re “looking at”</strong>. To accomplish this goal, I used a dataset containing 100,000 images of 43 types of labeled street signs to create supervised deep learning models that utilized CNNs and were capable of pretty reliably (more on that later) classifying new images of street signs. For a great summary of what all that means, check out this <a href="https://medium.com/intro-to-artificial-intelligence/deep-learning-series-1-intro-to-deep-learning-abb1780ee20">series of articles by Dhanoop Karunakaran</a>.</p><p>When discussing this type of technology, it is important to note the potential ramifications of a mistake. In the case of self-driving cars, a misidentified stop sign <em>or pedestrian</em> could easily cost a life. In a formal study, the significance of misclassifying each sign might be evaluated and given a weight. In this way, a new model could be trained to “err on the side of caution” when classifying more “critical” street signs. And while that all sounds great on paper, the ethics behind these types of choices must also be evaluated seriously. If our weighted inputs caused our new model to make fewer mistakes when classifying stop signs, for example, but performed slightly worse when classifying pedestrian crossing signs, would that be “acceptable”? With all that said, for the purposes of this post, we’ll focus more on the technology itself.</p><p>So, just how do you teach a computer how to identify different objects? To touch on that, we’ll begin with a simple example. As humans, when we look at the picture below, we see a cute dog with black and grey fur standing on a speckled rug.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IWPyQoZVxZQ_2kFp07kSGw.jpeg" /></figure><p>Without thought, we know that the subject of the image isn’t a horse, car, or cat. Even when viewing an example of a heavily edited photo I took of the same dog, as humans we still have no difficulty in identifying that the subject of the image is a dog, even though he is blue. Take a look:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*wmPcxOuhrrV89-3Lls_pTw.jpeg" /></figure><p>Computers, unsurprisingly, have no notion of what a dog is, much less what one looks like. Instead, computers see images as grouped matrices called <strong>tensors</strong>. In each matrix, each number corresponds to a specific pixel in the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*P9cVNLe1mcLQOO_m9gjcYA.gif" /></figure><p>Using convolutional neural networks, we are able to train models to extract image features such as edges, shapes, and colors for use in image classification. Using supervised learning models, we can pass thousands of images of dogs through a model in order to “teach the computer” what a dog looks like. So, while a computer may not appreciate how cute my dog Oliver is, by applying different filters to his portrait, it can determine that Oliver’s facial structure, color, and shape share similarities with other dogs, and that he is very likely to be a dog as well.</p><p>The same techniques described can be used to classify any number of target classes, like street signs. To illustrate this process, we’ll use an extremely “light” (read: weak) CNN that I created using the <a href="https://keras.io/">Keras</a> library in Python as an example to explain a bit more about what’s going on “under the hood” during the process of image classification. If you’d like to see the code I used to generate these examples, or view examples of more robust CNNs I trained, as well as take a look at some of the other processes required to train CNNs, such as image pre-processing, exploratory data analysis, etc, be sure to check out my code <a href="https://github.com/pawoody/DL_street_signs/blob/master/traffic_signs_modeling.ipynb">here</a>.</p><p>In this post, however, we’ll use the very basic CNN summarized by the image below to shed some light into the way a computer is able to “see” and “learn”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/527/1*_J_SvJME1PQSpgxpHW48Ow.png" /></figure><p>So, what does this mean? Why are we doing it and how does it help us to teach a computer to classify street signs?</p><p>First, we have our input images, all with dimensions of 32x32x3, or more simply, images that are 32 pixels in height and width, with 3 color channels (red, green, and blue). Because these inputs differ from those of other neural networks, where model inputs are vectors, we’ll need to massage our input images in order to reliably classify them using Deep Learning. For this example, I selected a random image from the dataset. Let’s take a look at the example, as well as display how the image is evaluated by our model:</p><pre># Selecting an example image<br>example = plt.imshow(data[&#39;x_train&#39;][2, :, :, :])</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*GqMthZcRXsbt_yhXQME2uQ.jpeg" /></figure><p>As I mentioned previously, computers “see” images as grouped matrices called tensors. Let’s take a look using the same example image:</p><pre># Displaying Image Tensor<br>how_the_computer_sees_it = data[&#39;x_train&#39;][2]<br>display(how_the_computer_sees_it)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kwJ9JrV5CU6imeeGAelG6Q.png" /></figure><p>Now that we’ve taken a look at the model inputs, let’s take a closer look at the process:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dQl4axV4-MSEbCCbomRq4Q.png" /></figure><p>Following the figure above, the first layer applied to the input image is a convolutional layer. In this step, we apply a convolutional kernel to our image, effectively sliding a smaller 3x3 pixel filter over the input image, evaluating their dot products. In the case above, we’ve chosen to apply 32 filters in this convolutional step, resizing the dimensionality of the layer output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*2iP-Z-YU-xrEP2sn6yRRmA.png" /></figure><p>Next, a pooling layer is applied to reduce the dimensionality (and therefore parameters) of the model using a sliding 2x2 filter, selecting the maximum value for each “window” or “chunk” of our parsed image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*DNuENzZRnf_7HCJcS0uxhQ.png" /></figure><p>Finally, in order to massage our image input into a vectorized format compatible with the dense layers of our neural network we’ll use to classify our images, we flatten the image input into an array.</p><p>While we can see the highlighting of edges and shapes in the image above, our outputs don’t appear very “clear” to the human eye. Despite this, even this very simple CNN classified images of street signs from the database relatively well (~80%). Following the creation and evaluation of this baseline model, I trained additional, tuned CNNs and evaluated their performance with an accuracy of &gt;94%. You can check out my code and conclusions by checking out this <a href="https://github.com/pawoody/DL_street_signs/blob/master/traffic_signs_modeling.ipynb">link</a>.</p><p>In a real world context, the task at hand is much more complex than simply classifying signs. In the context of self-driving, objects of many types, from many angles, must all be processed, classified, and acted upon extremely reliably at high speeds, all with human lives hanging in the balance.</p><p>Among of the most significant factors to remember when considering the results of our experiment, and others that use neural networks to classify image data, is the image quality used to train our models. While using very small images greatly reduces the time/resources required to train deep learning models, it can also very likely hamper classification performance. This same challenge is amplified greatly when models must account for many more factors all while controlling a vehicle moving at high speeds. This detail highlights the immense challenges faced by companies like Tesla, and the significance of their recent acquisition of <a href="https://fortune.com/2019/10/02/tesla-autopilot-ai-deepscale/">DeepScale</a>, a company focused on reducing the resources required to classify image data obtained from sensors in self-driving cars.</p><p><em>Originally published at </em><a href="https://medium.com/swlh/using-convolutional-neural-networks-to-classify-street-signs-a84b55e11e83"><em>https://medium.com</em></a><em> on November 25, 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eef7bd6a200a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/using-convolutional-neural-networks-to-classify-street-signs-eef7bd6a200a">Using Convolutional Neural Networks to Classify Street Signs</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Convolutional Neural Networks to Classify Street Signs]]></title>
            <link>https://medium.com/@paul.woody/using-convolutional-neural-networks-to-classify-street-signs-a953f9d7bdc9?source=rss-fac77cf82aef------2</link>
            <guid isPermaLink="false">https://medium.com/p/a953f9d7bdc9</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <category><![CDATA[convolutional-network]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Paul Woody]]></dc:creator>
            <pubDate>Mon, 25 Nov 2019 21:24:06 GMT</pubDate>
            <atom:updated>2019-11-26T16:08:43.829Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3bnhGykmfZPbwl5AZ2PLYQ.jpeg" /><figcaption>Image Source: yonah.org</figcaption></figure><p>Since the invention of the automobile, manufacturers have steadily added more safety features and improved car design over time with the goal of keeping drivers safer on the road. Automotive manufacturers have spent millions of dollars researching safety improvements for seatbelts, tires, and pretty much every car piece or part imaginable. Despite all of this investment, driving remains substantially more fatal than alternatives such as <a href="https://www.theguardian.com/world/2019/jan/02/plane-crash-deaths-jump-sharply-in-2018-but-fatalities-still-rare">air travel</a> in 2019. According to the <a href="https://www.nsc.org/road-safety/safety-topics/fatality-estimates">National Safety Council</a>, approximately 40,000 people died in automotive accidents in the United States alone in 2018. In fact, there were a total of ~500 deaths resulting from plane crashes recorded globally in 2018 — that’s <strong>80 times</strong> <strong>fewer deaths</strong> when compared to car crash fatalities in the US only.</p><p>There are many factors that likely contribute to this drastic difference, such as the difficulty in training and certification for commercial pilots, compared to motor vehicle drivers and the amount of cars on the road vs. planes. Despite this, the difference in mortality remains, in my opinion, pretty staggering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ajw3nluAdCp3lJ9n.jpeg" /></figure><blockquote>Today, companies such as Tesla, Volvo, and many others are investing into technology such as Artificial Intelligence that seeks to correct the root cause of the issue, and the one we’ve previously ignored — <strong>human performance behind the wheel.</strong></blockquote><p>Perhaps the most exciting advancement in automotive technology since the invention of the car itself is the birth of self-driving, or autonomous vehicles. Once the topic of Science Fiction, self-driving cars have been made possible through the power of Convolutional Neural Networks (CNNs) and other modeling techniques. But, before we hand over the keys entirely, there are 3 major challenges we must overcome.</p><p>We must:</p><p>&gt; 1. Obtain the data (sensors, cameras)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/1*juuc_tJPtezMivSYcqH-gg.jpeg" /><figcaption>Image source: <a href="https://www.sae.org/*">www.sae.org</a></figcaption></figure><p>&gt; 2. Process the data (AI — we’ll focus on this part)</p><p>&gt; 3. Act on the data (drive the car)</p><p>To highlight a very small sliver of the technological challenges faced by companies like Tesla in bringing autonomous cars to the public, we’ll focus exclusively on a simplified application of step 2, <strong>processing image data in order to determine what it is we’re “looking at”</strong>. To accomplish this goal, I used a dataset containing 100,000 images of 43 types of labeled street signs to create supervised deep learning models that utilized CNNs and were capable of pretty reliably (more on that later) classifying new images of street signs. For a great summary of what all that means, check out this <a href="https://medium.com/intro-to-artificial-intelligence/deep-learning-series-1-intro-to-deep-learning-abb1780ee20">series of articles by Dhanoop Karunakaran</a>.</p><p>When discussing this type of technology, it is important to note the potential ramifications of a mistake. In the case of self-driving cars, a misidentified stop sign <em>or pedestrian</em> could easily cost a life. In a formal study, the significance of misclassifying each sign might be evaluated and given a weight. In this way, a new model could be trained to “err on the side of caution” when classifying more “critical” street signs. And while that all sounds great on paper, the ethics behind these types of choices must also be evaluated seriously. If our weighted inputs caused our new model to make fewer mistakes when classifying stop signs, for example, but performed slightly worse when classifying pedestrian crossing signs, would that be “acceptable”? With all that said, for the purposes of this post, we’ll focus more on the technology itself.</p><p>So, just how do you teach a computer how to identify different objects? To touch on that, we’ll begin with a simple example. As humans, when we look at the picture below, we see a cute dog with black and grey fur standing on a speckled rug.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IWPyQoZVxZQ_2kFp07kSGw.jpeg" /></figure><p>Without thought, we know that the subject of the image isn’t a horse, car, or cat. Even when viewing an example of a heavily edited photo I took of the same dog, as humans we still have no difficulty in identifying that the subject of the image is a dog, even though he is blue. Take a look:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*wmPcxOuhrrV89-3Lls_pTw.jpeg" /></figure><p>Computers, unsurprisingly, have no notion of what a dog is, much less what one looks like. Instead, computers see images as grouped matrices called <strong>tensors</strong>. In each matrix, each number corresponds to a specific pixel in the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*P9cVNLe1mcLQOO_m9gjcYA.gif" /></figure><p>Using convolutional neural networks, we are able to train models to extract image features such as edges, shapes, and colors for use in image classification. Using supervised learning models, we can pass thousands of images of dogs through a model in order to “teach the computer” what a dog looks like. So, while a computer may not appreciate how cute my dog Oliver is, by applying different filters to his portrait, it can determine that Oliver’s facial structure, color, and shape share similarities with other dogs, and that he is very likely to be a dog as well.</p><p>The same techniques described can be used to classify any number of target classes, like street signs. To illustrate this process, we’ll use an extremely “light” (read: weak) CNN that I created using the <a href="https://keras.io/">Keras</a> library in Python as an example to explain a bit more about what’s going on “under the hood” during the process of image classification. If you’d like to see the code I used to generate these examples, or view examples of more robust CNNs I trained, as well as take a look at some of the other processes required to train CNNs, such as image pre-processing, exploratory data analysis, etc, be sure to check out my code <a href="https://github.com/pawoody/DL_street_signs/blob/master/traffic_signs_modeling.ipynb">here</a>.</p><p>In this post, however, we’ll use the very basic CNN summarized by the image below to shed some light into the way a computer is able to “see” and “learn”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/527/1*_J_SvJME1PQSpgxpHW48Ow.png" /></figure><p>So, what does this mean? Why are we doing it and how does it help us to teach a computer to classify street signs?</p><p>First, we have our input images, all with dimensions of 32x32x3, or more simply, images that are 32 pixels in height and width, with 3 color channels (red, green, and blue). Because these inputs differ from those of other neural networks, where model inputs are vectors, we’ll need to massage our input images in order to reliably classify them using Deep Learning. For this example, I selected a random image from the dataset. Let’s take a look at the example, as well as display how the image is evaluated by our model:</p><pre># Selecting an example image<br>example = plt.imshow(data[&#39;x_train&#39;][2, :, :, :])</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*GqMthZcRXsbt_yhXQME2uQ.jpeg" /><figcaption>How Humans see it</figcaption></figure><p>As I mentioned previously, computers “see” images as grouped matrices called tensors. Let’s take a look using the same example image:</p><pre># Displaying Image Tensor<br>how_the_computer_sees_it = data[&#39;x_train&#39;][2]<br>display(how_the_computer_sees_it)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kwJ9JrV5CU6imeeGAelG6Q.png" /></figure><p>Now that we’ve taken a look at the model inputs, let’s take a closer look at the process:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dQl4axV4-MSEbCCbomRq4Q.png" /></figure><p>Following the figure above, the first layer applied to the input image is a convolutional layer. In this step, we apply a convolutional kernel to our image, effectively sliding a smaller 3x3 pixel filter over the input image, evaluating their dot products. In the case above, we’ve chosen to apply 32 filters in this convolutional step, resizing the dimensionality of the layer output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*2iP-Z-YU-xrEP2sn6yRRmA.png" /></figure><p>Next, a pooling layer is applied to reduce the dimensionality (and therefore parameters) of the model using a sliding 2x2 filter, selecting the maximum value for each “window” or “chunk” of our parsed image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*DNuENzZRnf_7HCJcS0uxhQ.png" /></figure><p>Finally, in order to massage our image input into a vectorized format compatible with the dense layers of our neural network we’ll use to classify our images, we flatten the image input into an array.</p><p>While we can see the highlighting of edges and shapes in the image above, our outputs don’t appear very “clear” to the human eye. Despite this, even this very simple CNN classified images of street signs from the database relatively well (~80%). Following the creation and evaluation of this baseline model, I trained additional, tuned CNNs and evaluated their performance with an accuracy of &gt;94%. You can check out my code and conclusions by checking out this <a href="https://github.com/pawoody/DL_street_signs/blob/master/traffic_signs_modeling.ipynb">link</a>.</p><p>In a real world context, the task at hand is much more complex than simply classifying signs. In the context of self-driving, objects of many types, from many angles, must all be processed, classified, and acted upon extremely reliably at high speeds, all with human lives hanging in the balance.</p><p>Among of the most significant factors to remember when considering the results of our experiment, and others that use neural networks to classify image data, is the image quality used to train our models. While using very small images greatly reduces the time/resources required to train deep learning models, it can also very likely hamper classification performance. This same challenge is amplified greatly when models must account for many more factors all while controlling a vehicle moving at high speeds. This detail highlights the immense challenges faced by companies like Tesla, and the significance of their recent acquisition of <a href="https://fortune.com/2019/10/02/tesla-autopilot-ai-deepscale/">DeepScale</a>, a company focused on reducing the resources required to classify image data obtained from sensors in self-driving cars.</p><p><em>Originally published at </em><a href="https://medium.com/swlh/using-convolutional-neural-networks-to-classify-street-signs-a84b55e11e83"><em>https://medium.com</em></a><em> on November 25, 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a953f9d7bdc9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Convolutional Neural Networks to Classify Street Signs]]></title>
            <link>https://medium.com/swlh/using-convolutional-neural-networks-to-classify-street-signs-a84b55e11e83?source=rss-fac77cf82aef------2</link>
            <guid isPermaLink="false">https://medium.com/p/a84b55e11e83</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[supervised-learning]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Paul Woody]]></dc:creator>
            <pubDate>Sat, 23 Nov 2019 21:09:02 GMT</pubDate>
            <atom:updated>2019-11-25T21:24:06.324Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>A glimpse into the technology used in driver-less cars</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3bnhGykmfZPbwl5AZ2PLYQ.jpeg" /><figcaption>Image Source: yonah.org</figcaption></figure><p>Since the invention of the automobile, manufacturers have steadily added more safety features and improved car design over time with the goal of keeping drivers safer on the road. Automotive manufacturers have spent millions of dollars researching safety improvements for seatbelts, tires, and pretty much every car piece or part imaginable. Despite all of this investment, driving remains substantially more fatal than alternatives such as <a href="https://www.theguardian.com/world/2019/jan/02/plane-crash-deaths-jump-sharply-in-2018-but-fatalities-still-rare">air travel</a> in 2019. According to the <a href="https://www.nsc.org/road-safety/safety-topics/fatality-estimates">National Safety Council</a>, approximately 40,000 people died in automotive accidents in the United States alone in 2018. In fact, there were a total of ~500 deaths resulting from plane crashes recorded globally in 2018 — that’s <strong>80 times</strong> <strong>fewer deaths</strong> when compared to car crash fatalities in the US only.</p><p>There are many factors that likely contribute to this drastic difference, such as the difficulty in training and certification for commercial pilots, compared to motor vehicle drivers and the amount of cars on the road vs. planes. Despite this, the difference in mortality remains, in my opinion, pretty staggering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BDWKc4yo8AI0xL7rlIXZKA.jpeg" /></figure><blockquote>Today, companies such as Tesla, Volvo, and many others are investing into technology such as Artificial Intelligence that seeks to correct the root cause of the issue, and the one we’ve previously ignored — <strong>human performance behind the wheel.</strong></blockquote><p>Perhaps the most exciting advancement in automotive technology since the invention of the car itself is the birth of self-driving, or autonomous vehicles. Once the topic of Science Fiction, self-driving cars have been made possible through the power of Convolutional Neural Networks (CNNs) and other modeling techniques. But, before we hand over the keys entirely, there are 3 major challenges we must overcome.</p><p>We must:</p><p>&gt; 1. Obtain the data (sensors, cameras)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/1*juuc_tJPtezMivSYcqH-gg.jpeg" /><figcaption>Image source: <a href="https://www.sae.org/*">https://www.sae.org/</a></figcaption></figure><p>&gt; 2. Process the data (AI — we’ll focus on this part)</p><p>&gt; 3. Act on the data (drive the car)</p><p>To highlight a very small sliver of the technological challenges faced by companies like Tesla in bringing autonomous cars to the public, we’ll focus exclusively on a simplified application of step 2, <strong>processing image data in order to determine what it is we’re “looking at”</strong>. To accomplish this goal, I used a dataset containing 100,000 images of 43 types of labeled street signs to create supervised deep learning models that utilized CNNs and were capable of pretty reliably (more on that later) classifying new images of street signs. For a great summary of what all that means, check out this <a href="https://medium.com/intro-to-artificial-intelligence/deep-learning-series-1-intro-to-deep-learning-abb1780ee20">series of articles by Dhanoop Karunakaran</a>.</p><p>When discussing this type of technology, it is important to note the potential ramifications of a mistake. In the case of self-driving cars, a misidentified stop sign <em>or pedestrian</em> could easily cost a life. In a formal study, the significance of misclassifying each sign might be evaluated and given a weight. In this way, a new model could be trained to “err on the side of caution” when classifying more “critical” street signs. And while that all sounds great on paper, the ethics behind these types of choices must also be evaluated seriously. If our weighted inputs caused our new model to make fewer mistakes when classifying stop signs, for example, but performed slightly worse when classifying pedestrian crossing signs, would that be “acceptable”? With all that said, for the purposes of this post, we’ll focus more on the technology itself.</p><p>So, just how do you teach a computer how to identify different objects? To touch on that, we’ll begin with a simple example. As humans, when we look at the picture below, we see a cute dog with black and grey fur standing on a speckled rug.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IWPyQoZVxZQ_2kFp07kSGw.jpeg" /></figure><p>Without thought, we know that the subject of the image isn’t a horse, car, or cat. Even when viewing an example of a heavily edited photo I took of the same dog, as humans we still have no difficulty in identifying that the subject of the image is a dog, even though he is blue. Take a look:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*wmPcxOuhrrV89-3Lls_pTw.jpeg" /></figure><p>Computers, unsurprisingly, have no notion of what a dog is, much less what one looks like. Instead, computers see images as grouped matrices called <strong>tensors</strong>. In each matrix, each number corresponds to a specific pixel in the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*P9cVNLe1mcLQOO_m9gjcYA.gif" /></figure><p>Using convolutional neural networks, we are able to train models to extract image features such as edges, shapes, and colors for use in image classification. Using supervised learning models, we can pass thousands of images of dogs through a model in order to “teach the computer” what a dog looks like. So, while a computer may not appreciate how cute my dog Oliver is, by applying different filters to his portrait, it can determine that Oliver’s facial structure, color, and shape share similarities with other dogs, and that he is very likely to be a dog as well.</p><p>The same techniques described can be used to classify any number of target classes, like street signs. To illustrate this process, we’ll use an extremely “light” (read: weak) CNN that I created using the <a href="https://keras.io/">Keras</a> library in Python as an example to explain a bit more about what’s going on “under the hood” during the process of image classification. If you’d like to see the code I used to generate these examples, or view examples of more robust CNNs I trained, as well as take a look at some of the other processes required to train CNNs, such as image pre-processing, exploratory data analysis, etc, be sure to check out my code <a href="https://github.com/pawoody/DL_street_signs/blob/master/traffic_signs_modeling.ipynb">here</a>.</p><p>In this post, however, we’ll use the very basic CNN summarized by the image below to shed some light into the way a computer is able to “see” and “learn”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/527/1*_J_SvJME1PQSpgxpHW48Ow.png" /></figure><p>So, what does this mean? Why are we doing it and how does it help us to teach a computer to classify street signs?</p><p>First, we have our input images, all with dimensions of 32x32x3, or more simply, images that are 32 pixels in height and width, with 3 color channels (red, green, and blue). Because these inputs differ from those of other neural networks, where model inputs are vectors, we’ll need to massage our input images in order to reliably classify them using Deep Learning. For this example, I selected a random image from the dataset. Let’s take a look at the example, as well as display how the image is evaluated by our model:</p><pre># Selecting an example image<br>example = plt.imshow(data[&#39;x_train&#39;][2, :, :, :])</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*GqMthZcRXsbt_yhXQME2uQ.jpeg" /><figcaption>How humans see it</figcaption></figure><p>As I mentioned previously, computers “see” images as grouped matrices called tensors. Let’s take a look using the same example image:</p><pre># Displaying Image Tensor<br>how_the_computer_sees_it = data[&#39;x_train&#39;][2]<br>display(how_the_computer_sees_it)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kwJ9JrV5CU6imeeGAelG6Q.png" /></figure><p>Now that we’ve taken a look at the model inputs, let’s take a closer look at the process:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dQl4axV4-MSEbCCbomRq4Q.png" /></figure><p>Following the figure above, the first layer applied to the input image is a convolutional layer. In this step, we apply a convolutional kernel to our image, effectively sliding a smaller 3x3 pixel filter over the input image, evaluating their dot products. In the case above, we’ve chosen to apply 32 filters in this convolutional step, resizing the dimensionality of the layer output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*2iP-Z-YU-xrEP2sn6yRRmA.png" /></figure><p>Next, a pooling layer is applied to reduce the dimensionality (and therefore parameters) of the model using a sliding 2x2 filter, selecting the maximum value for each “window” or “chunk” of our parsed image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*DNuENzZRnf_7HCJcS0uxhQ.png" /></figure><p>Finally, in order to massage our image input into a vectorized format compatible with the dense layers of our neural network we’ll use to classify our images, we flatten the image input into an array.</p><p>While we can see the highlighting of edges and shapes in the image above, our outputs don’t appear very “clear” to the human eye. Despite this, even this very simple CNN classified images of street signs from the database relatively well (~80%). Following the creation and evaluation of this baseline model, I trained additional, tuned CNNs and evaluated their performance with an accuracy of &gt;94%. You can check out my code and conclusions by checking out this <a href="https://github.com/pawoody/DL_street_signs/blob/master/traffic_signs_modeling.ipynb">link</a>.</p><p>In a real world context, the task at hand is much more complex than simply classifying signs. In the context of self-driving, objects of many types, from many angles, must all be processed, classified, and acted upon extremely reliably at high speeds, all with human lives hanging in the balance.</p><p>Among of the most significant factors to remember when considering the results of our experiment, and others that use neural networks to classify image data, is the image quality used to train our models. While using very small images greatly reduces the time/resources required to train deep learning models, it can also very likely hamper classification performance. This same challenge is amplified greatly when models must account for many more factors all while controlling a vehicle moving at high speeds. This detail highlights the immense challenges faced by companies like Tesla, and the significance of their recent acquisition of <a href="https://fortune.com/2019/10/02/tesla-autopilot-ai-deepscale/">DeepScale</a>, a company focused on reducing the resources required to classify image data obtained from sensors in self-driving cars.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a84b55e11e83" width="1" height="1" alt=""><hr><p><a href="https://medium.com/swlh/using-convolutional-neural-networks-to-classify-street-signs-a84b55e11e83">Using Convolutional Neural Networks to Classify Street Signs</a> was originally published in <a href="https://medium.com/swlh">The Startup</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Power of Data — A Review of MLconf 2019]]></title>
            <link>https://medium.datadriveninvestor.com/the-power-of-data-a-review-of-mlconf-2019-f720124de7b7?source=rss-fac77cf82aef------2</link>
            <guid isPermaLink="false">https://medium.com/p/f720124de7b7</guid>
            <category><![CDATA[mlconf]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Paul Woody]]></dc:creator>
            <pubDate>Tue, 12 Nov 2019 23:42:18 GMT</pubDate>
            <atom:updated>2019-11-27T11:39:01.802Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Power of Data — A Review of MLconf 2019</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NajFK2U5gNcSI2zYpl330g.jpeg" /><figcaption>MLconf San Francisco — 2019</figcaption></figure><p>(983 words)</p><p>As I laced up my shoes and began my morning walk to MLconf last Friday, “Fog City” lived up to its namesake. In sharp contrast to the chilly November weather, the content at MLconf San Francisco was anything but. The event, hosted near Union Square, boasted a diverse panel of speakers from big-name companies such as Uber, Netflix, Google, Airbnb, Amazon, and many others, all pushing the bounds of what is “possible” through the use of AI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KFt8IU2UwbGOFWIQCxgzdg.jpeg" /></figure><p>Having myself committed this year to begin a lifelong journey of learning and applying ML in my own career, I was beyond excited to have the opportunity to attend.</p><p>I wasn’t disappointed.</p><p>Data Science is improving human understanding of the world around us in ways we could have never imagined even just years ago. This incredible diversity in application was evident, as topics from presenters ranged from predicting the onset of Alzheimer’s disease based on speech to the technology utilized to minimize the time you and I wait for our Ubers to arrive. Despite the diverse applications and industries represented at the conference, one single truth rings clearest — AI impacts our daily lives more than most realize, and it is critical that we do all that we can to seek to understand, and ethically implement AI to benefit our lives and businesses.</p><p>While each presentation from the event undoubtably deserves a post of its own, I wanted to highlight a few of the talks that I found compelling, as well as share links for anyone interested in learning more.</p><p>Let’s get started!</p><p>Kicking off the event was <a href="https://www.linkedin.com/in/franziska-bell-phd-00097992/">Dr. Franziska Bell</a>, Director of Data Science at Uber. In her talk, Dr. Bell discussed some of the ways that Uber tackles a commonly faced challenge — model reproducibility and ability to collaborate and share the results of deep learning models at scale. It’s no secret that Uber’s growth has been explosive since it was founded just 10 years ago. With rapid growth, comes the need for implementation of scalable process and technology to support it.</p><p>A primary factor in the success of rideshare companies like Uber is convenience. In most places in the country and many around the globe (ask me about my adventures using Uber in Kyiv), you can take out your phone and reliably be en route to your next destination within 5–10 minutes. As millions of users continue to join the platform worldwide, the challenge of maintaining the standards of convenience and reliability that Uber’s customers have come to expect becomes increasingly difficult to manage. It’s a challenge of scale that Uber tackles using Machine Learning.</p><p><a href="https://www.datadriveninvestor.com/2019/03/03/editors-pick-5-machine-learning-books/">DDI Editor&#39;s Pick: 5 Machine Learning Books That Turn You from Novice to Expert | Data Driven Investor</a></p><p>If you’ve ever spent time training predictive models, you are most likely familiar with the pains presented by the limitations in power of desktop machines and storing/sharing modeling results with other members of a team. Enter <a href="https://eng.uber.com/michelangelo/">Michelangelo</a>, an Uber in-house solution designed to address the following challenges Uber faces in order to put their data to use:</p><ol><li>Manage data</li><li>Train models</li><li>Evaluate models</li><li>Deploy models</li><li>Make predictions</li><li>Monitor predictions</li></ol><p>Alongside Michelangelo, Dr. Bell shared another Uber in-house tool used to build on the accessibility Michelagelo provides that Uber calls its “<a href="https://eng.uber.com/dsw/">Data Science Workbench</a>”. Adhering to their philosophy of “Nail it, then scale it”, the tool is a centralized platform for data exploration, preparation, modeling, and analysis, with the flexibility to accommodate users of varying skill levels. Uber uses these models, incorporating anticipated surges based on upcoming events, weather, time of day, and a plethora of other factors to predict demand and properly incentivize drivers to meet surging demands. You can read more about the work of Dr. Bell and her team at Uber <a href="https://eng.uber.com/">here</a>.</p><p>The challenge of accessibility of Machine Learning seemed to be a common one in talks by several of the speakers at the event. Thankfully (for many of us), it doesn’t require a PhD to understand and apply ML in business today. <a href="https://www.linkedin.com/in/mpei/">Martin Isaksson</a>, CEO &amp; Co-Founder of <a href="https://perceptilabs.readme.io/">PerceptiLabs</a>, shared his company’s vision to bring the power of ML to the masses; drag and drop neural networks. While not entirely “hands off”, the PerceptiLabs drag and drop platform is capable of constructing a fully-functional convolutional neural network (CNN) very simply, including handling image preprocessing and visualizing intermediate activations at each layer, allowing users of varying skill levels to quickly check their model outputs and debug issues along the way. Using (editable) boiler plate code, the platform allows users to connect layers of a CNN, as well as to complete training and analyze results. During a short live demo, Martin’s team showed the process of creating and evaluating the performance of a “basic” CNN from start to finish in just minutes, all without writing a single line of code.</p><p>But it’s not just convenience, customer success, and accessibility. Machine Learning is being put to use for groundbreaking medical research as well. In her talk, <a href="https://www.linkedin.com/in/jnovikova/">Dr. Jekaterina Novikova</a> discussed how her research at <a href="https://winterlightlabs.com/">Winterlight Labs</a> is combating Alzheimer’s Disease, a condition affecting millions worldwide. Using automatic speech recognition and ML models based on speech patterns, Dr. Novikova’s model quantifies speech, language, and cognition in order to classify normal speech from that of individuals in early development of Alzheimer’s disease, thus improving rates of early diagnosis and patient prognosis. For those interested in learning more, you can check out more of Dr. Novikova’s work on her <a href="https://jeknov.tumblr.com/works">blog</a>.</p><p>In summary, if you are a student, data scientist, data engineer, business owner, or just someone interested in learning about and meeting some of the folks on the cutting edge of the world of Artificial Intelligence, you simply can’t miss the action at MLconf.</p><p><em>For more information on the speakers at MLconf San Francisco 2019, you can check out this </em><a href="https://mlconf.com/speakers/"><em>link</em></a><em>. Special thanks to </em><a href="https://www.linkedin.com/in/burtoncourtney/"><em>Courtney Burton</em></a><em>, Founder of MLconf, and the rest of her team for organizing this awesome event!</em></p><p>Hope to see you all at the next one!</p><p>Contact: <a href="https://www.linkedin.com/in/paul-woody/">https://www.linkedin.com/in/paul-woody/</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fb2a0d6%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2Fb2a0d6%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href">https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f720124de7b7" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/the-power-of-data-a-review-of-mlconf-2019-f720124de7b7">The Power of Data — A Review of MLconf 2019</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>