Google Data

A few weeks back, we ran a poll on our Google+ page asking what type of content you all would be interested in seeing more of on our pages. One of the top responses was information on the interviewing and hiring process. In turn, we decided that we should write a series of posts shedding some light on how hiring works at Google. This entry will primarily cover what components our interviews are comprised of, but if you want to dive deeper into our hiring practices and many other aspects of Google, be sure to check out Work Rules! by Laszlo Bock, who heads up People Operations at Google.

As with just about everything at Google, our interviews are based on—and constantly improved upon—by data. This reliance on data is of paramount importance, because hiring has the ability to either make or break a company. Plus, research has shown that if data is left out of the hiring equation, the decision of who should be hired and who shouldn’t is often swayed by an interviewer’s unconscious biases and superficial snap judgments. Consequently, according to Laszlo Bock, “Most interviews are a waste of time because 99.4% of the time is spent trying to confirm whatever impression the interviewer formed in the first ten seconds.” Fortunately, Google interviews are not like most interviews. But, how exactly are they different?

To begin, Google interviews only include question formats that have been proven to actually predict job performance. In 1998, Frank Schmidt and John Hunter published a meta-analysis of 85 years of research on how well different types of assessments predict job performance. Overall, they identified three assessment techniques that were more effective than all others: work sample tests, tests of general cognitive ability, and structured interviews. As you may have guessed, we incorporate all three within our interview process.

Work sample tests entail giving a candidate a sample piece of work or problem to complete that is similar to what the candidate would face on the job. At Google, all our technical hires (engineering and product management) are tasked with a work sample test of sorts, where they are asked to solve engineering problems during the interview. This allows candidates to showcase their skills, while also giving us a chance to see how they go about attacking actual problems.

Next, tests of general cognitive ability involve measuring raw intelligence and the ability to problem solve, reason, and learn. However, in contrast to case interviews and brainteasers (neither of which is used by Google), these tests have defined right and wrong answers. For this measure, we want smart people who can learn and adapt to new situations. In general, Google interview questions testing general cognitive ability try to get at how candidates have solved hard problems in real life and how they learn.

Last but not least, structured interviews ask a consistent set of questions and have clear criteria for gauging the quality of responses to these questions. Structured interviews help isolate candidate performance from other variables, so that candidates are all judged equally, no matter who interviewed them. Altogether, there are two kinds of structured interviews: behavioral and situational. Behavioral interviews ask candidates to describe prior achievements (“Tell me about a time…”), while situational interviews present a job-related hypothetical situation (“What would you do if…”). Over the course of a Google interview, interviewers ask both types of questions.

So, there it is…a quick glimpse into the components that make up a Google interview. All in all, the primary objective of our interviews is to accurately predict how candidates would perform if they joined the team. By making sure that we use only those interview formats that have been proven to best predict future job performance and evaluate each candidate equally, we have been able to maintain our high hiring standards, while simultaneously making the interview process more fair and rewarding for candidates.

Posted by Steven Claunch, Online Hiring and Insights Team

Posted by Andrea Held, Google University Relations & Matt Cooke, Google News Lab Europe

Journalism is evolving fast in the digital age, and researchers across Europe are working on exciting projects to create innovative new tools and open source software that will support online journalism and benefit readers. As part of the wider Google Digital News Initiative (DNI), we invited academic researchers across Europe to submit proposals for the Computational Journalism Research Awards.

After careful review by Google’s News Lab and Research teams, the following projects were selected:

SCAN: Systematic Content Analysis of User Comments for Journalists
Walid Maalej, Professor of Informatics, University of Hamburg
Wiebke Loosen, Senior Researcher for Journalism, Hans-Bredow-Institute, Hamburg, Germany
This project aims at developing a framework for the systematic, semi-automated analysis of audience feedback on journalistic content to better reflect the voice of users, mitigate the analysis efforts, and help journalists generate new content from the user comments.

Event Thread Extraction for Viewpoint Analysis
Ioana Manolescu, Senior Researcher, INRIA Saclay, France
Xavier Tannier, Professor of Computer Science, University Paris-Sud, France
The goal of the project is to automatically build topic “event threads” that will help journalists and citizens decode claims made by public figures, in order to distinguish between personal opinion, communication tools and voluntary distortions of the reality.

Computational Support for Creative Story Development by Journalists
Neil Maiden, Professor of Systems Engineering
George Brock, Professor of Journalism, City University London, UK
This project will develop a new software prototype to implement creative search strategies that journalists could use to strengthen investigative storytelling more efficiently than with current news content management and search tools.

We congratulate the recipients of these awards and we look forward to the results of their research. Each award includes funding of up to $60,000 in cash and $20,000 in computing credits on Google’s Cloud Platform. Stay tuned for updates on their progress.

During the Translate Community: Google I/O Challenge nearly 2.5 million phrases or about 12 million words were translated and validated by participants during Google I/O and beyond. From Afrikaans to Zulu, we saw approximately 75,000 people representin…

In May, we added the following features to DoubleClick Search (DS):

• Support for AdWords “Mobile app installs” campaigns: Use DS to create, target, measure, and optimize on “Mobile app installs” campaigns, a new, simple type of AdWords campaign that focuses on getting people to download your app. Ads from these campaigns appear only on phones or tablets, and they land directly on your app page in the Google Play Store or Apple App Store. Learn more.

• Manage and report on local inventory ads: Local inventory ads advertise products that are available in nearby stores when a customer searches for items on Google. You can now use DS to manage and report on your local inventory ads along with online ads in AdWords Shopping campaigns. Learn more.

• New reporting column for AdWords campaigns: The Invalid clicks column displays the number of clicks on ads that Google considers to be illegitimate, such as unintentional clicks or clicks resulting from malicious software, as reported by AdWords.

• Ad rotation settings for Bing Ads: If you create more than one ad in a Bing Ads ad group, Bing Ads can either rotate through the ads evenly or predominantly show ads that have the highest click-through rate (CTR). You can use the DS UI or DS bulksheets to specify how Bing Ads should rotate ads in an ad group.

We also updated the following features:

• DS URL templates are now hierarchical: Use DS URL templates to apply a redirect URL or a set of URL parameters to landing page URLs for keywords and sitelinks. You can create DS URL templates at multiple levels within an engine account. For example, you can create separate URL templates for each campaign or ad group. Learn more.

• Improved support for Merchant Center multi-client accounts: If you use DS to manage AdWords Shopping campaigns or inventory keyword campaigns and you have a parent Merchant Center accounts, known as a multi-client account, you only need to link the parent account to DS and AdWords. Learn more.

Posted by the DoubleClick Search team

After meeting with developers in Chicago last week, we paid a visit to the Chicago Department of Transportation who is currently using the Google Maps APIs to coordinate their public projects. We parked our bus in front of their building in downtown Chicago to let all passersby come into the bus and look at how the DOT is executing the great infrastructure changes happening in their city.

We had the pleasure of meeting with William Cheaks, First Deputy Commissioner, Chicago Department of Transportation and Larry Olszak, Director of IT, Chicago Department of Transportation, to talk to us about how they use Google Maps APIs to help manage the city of Chicago’s infrastructure projects.

We also had a great surprise at the event—our friends at grubHub stopped by to bring the crew some delicious donuts. It was sweet treat to brighten everyone’s day. Even better, they used the Google Maps APIs delivered to us!

We’re looking forward to seeing you at one of our future stops. Join us in McLean, Virginia at Hilton Worldwide Headquarters on June 22 from 7am to 10am for a tour of the bus.

Posted by Hena Haines, Associate Product Marketing Manager, Google Maps APIs

Pride is one of the world’s greatest celebrations of diversity, one that we’re excited to support every year. For 2015, we’re doing something a little different: we’ve created an online parade, #AndProud, so that people from around the world can celebrate Pride together.

To join the global parade, visit Androidify, where you can create your own Android character with a fun new Pride wardrobe. During the weekend of June 27-28 your character will party side-by-side with others from around the world in the online parade.

But the celebration doesn’t end there. While the virtual parade happens online, thousands of Googlers will hit the streets of San Francisco, London and New York to show their support in those citywide Pride festivals. Some of the best #AndProud characters will appear on big screens as part of Google’s pride floats in all three cities.

In addition to #AndProud and our floats in SF, London and NYC, we’re celebrating Pride in our offices around the world, and in all sorts of ways across our products. We’re excited to be able to extend the celebration and give people around the world a new way to share their Pride.

Hope to see you at the parade!

Posted by Eddie Kalletta and Rich Terry, #AndProud parade marchers

Posted by Maru Ahues, Media Developer Advocate

When it comes to TV, content is king. But to enjoy great content, you first need to find it. We created Android TV with that in mind: a truly smart TV should deliver interesting content to users. Today, EPIX® joins a growing list of apps that use the Android TV platform to make it easy to enjoy movies, TV shows, sports highlights, music videos and more.

Making TV Apps Searchable

Think of your favorite movie. Now try to locate it in one of your streaming apps. If you have a few apps to choose from, it might take some hunting before you can watch that movie. With Android TV, we want to make it easier to be entertained. Finding ‘Teenage Mutant Ninja Turtles’ should be as easy as picking up the remote, saying ‘Teenage Mutant Ninja Turtles’ and letting the TV find it.

Searching for ‘Teenage Mutant Ninja Turtles’ shows results from Google Play and EPIX

You can drive users directly to content within your app by making it searchable from the Android TV search interface. Join app developers like EPIX, Sky News, YouTube, and Hulu Plus who are already making content discovery a breeze.

Recommending TV Content

When users want suggestions for content, the recommendations row on Android TV helps them quickly access relevant content right from the home screen. Recommendations are based on the user’s recent and frequent usage behaviors, as well as content preferences.

Recommendations from installed apps, like EPIX, appear in the Android TV home screen

Android TV allows developers to create recommendations for movies, TV shows, music and other types of content. Your app can provide recommendations to users to help get your content noticed. As an example, EPIX shows hollywood movies. NBA Game Time serves up basketball highlights. Washington Post offers video summaries of world events, and YouTube suggests videos based on your subscriptions and viewing history.

With less than one year since the consumer launch of Android TV, we’re already building upon a simpler, smarter and more personalized TV experience, and we can’t wait to see what you create.

After an amazing day with Harley-Davidson riders in Milwaukee, we made it to Chicago for our second developer meetup. We had a blast showing the developers the Code the Road bus in the middle of downtown Chicago. We hope all who attended had a fantastic time.

Before the event even started we hitched a ride with a few maps developers taking Lyfts into the Chicago office. We spoke to their drivers about topics ranging from the future of maps to who they wish they could carpool with in a Lyft. Once at the event, attendees had the opportunity to explore the bus, pick up pegman stickers (complete with the Lyft mustache), and meet other developers before heading up to hear four technical talks.

I was up first (traveling all the way from Sydney) to discuss using Google Maps APIs in Android apps. I outlined ways to use custom markers, colors, and animations to make maps more focused and easier for users to navigate.

Next up was Ed Boiling, Google Maps for Work Solution Architect, who demonstrated how to use Google Cloud Platform to manage, serve, visualize , and analyze location data.

We, then, welcomed Patrick Skoglund Geospatial Lead from SADA Systems. He outlined how using the Google Maps toolset can help large companies solve complex problems.

We ended the night hearing from Kyle Madsen, Android developer at Lyft. Kyle gave a great overview of his experience with Google Maps APIs at Lyft. He described how the ease of implementation using Google Maps APIs helped them to focus on their core Lyft application while relying on Google Maps for the navigation components.

Next up, we’re on our way to New York to meet with Sesame Street and a few of our youngest fans. We’ll also be hosting our last developer meetup on June 18, so sign-up to join us in New York.

We look forward to seeing you in New York or one of our future stops on the road!

Posted by Florian Bertele, Product Manager, Google Maps APIs

Posted by David McClusky, Google Maps Solutions Architect

Editor’s Note: We are (literally) coding a mobile app on the road as part of our cross-country road trip, Code the Road. If you haven’t read our previous posts in the series, be sure to check out Hitting the Road with the Google Maps Android API and Store Your Locations in the Cloud with Firebase to get started. Follow along as we build our app on the road and read more from David McClusky, Maps Solutions Architect at Google.

We’re now in the third week of our Code the Road journey and we’ve had a great time meeting with developers and customers across the country. In this third post in our app blog series we will extend the capabilities of our mobile application by adding the ability to ‘check-in’ to places we visit along the way using the Google Places API for Android. If you haven’t read our earlier posts, checkout our posts on building a mobile map with the Google Maps Android API and connecting it to the cloud.

Our Code the Road bus is now visiting New York City, so we thought it would be useful to update our app and add the capability to view nearby places, select places and check-in, and then store those locations in the cloud for future reference. To do this we’ll integrate the Google Places API for Android.

We’ll start by utilizing the Place Picker—an out-of-the-box user interface that allows your app to quickly connect to Google’s database of hundreds of millions points of interest.

Prerequisites

Before you get started, make sure you have the latest version of Android Studio installed and your project includes the Google Play services component. In addition, you must have an API Key for Google Places configured with your app’s digital certificate. For more information on creating a Places API key read the documentation. You should have also completed the first two steps in our blog series, as we’ll be utilizing our existing connection to Firebase to store our check-ins in the cloud.

Adding the Place Picker to Your Application

The Place Picker provides a UI dialog that displays an interactive map and a list of nearby places. When you activate the Place Picker, it attempts to identify the your current location, and will also provide a list of alternative suggestions. We’ll add some code to our app so when we select a place, we’ll add a marker to the map and store the location in the cloud.

It’s very easy to add the Place Picker to your application, as Google handles all of the user interface elements for you.

The first thing we’ll do is add a button to our app on the action bar to launch the place picker. View the documentation for adding an action button.

To launch the Place Picker from our action button, we will construct an intent in the onOptionsItemSelected() function:

@Override
public boolean onOptionsItemSelected(MenuItem item) {
 // Handle presses on the action bar items
 switch (item.getItemId()) {
   case R.id.action_search:
     try {
        PlacePicker.IntentBuilder intentBuilder = new PlacePicker.IntentBuilder();
        Intent intent = intentBuilder.build(this);

        // Start the Intent by requesting a result, identified by a request code.
        startActivityForResult(intent, REQUEST_PLACE_PICKER);

     } catch (GooglePlayServicesRepairableException e) {
       GooglePlayServicesUtil.getErrorDialog(e.getConnectionStatusCode(), 
       this, 0);
     } catch (GooglePlayServicesNotAvailableException e) {
       Toast.makeText(this, "Google Play Services is not available.",
Toast.LENGTH_LONG).show();
     }
     return true;

  default:
    return super.onOptionsItemSelected(item);
  }
}

After we select a place, we’ll need to identify which location was selected and display a marker on the map. First we will capture the result by creating the onActivityResult() function. We first obtain a reference to the place selected:

final Place place = PlacePicker.getPlace(data, this);

And then extract the various details about the place:

final CharSequence name = place.getName();
final CharSequence address = place.getAddress();
final CharSequence phone = place.getPhoneNumber();

Finally we can add a marker to the map:

mMap.addMarker(new MarkerOptions()
                   .position(place.getLatLng())
                   .title(name.toString())

Here is what the complete function in our app looks like:

@Override
public void onActivityResult(int requestCode, int resultCode, Intent data) {

 if (requestCode == REQUEST_PLACE_PICKER) {
    // This result is from the PlacePicker dialog.

    if (resultCode == Activity.RESULT_OK) {
       /* User has picked a place, extract data.
          Data is extracted from the returned intent by retrieving a Place 
          object from the PlacePicker.
         */
         final Place place = PlacePicker.getPlace(data, this);

         /* A Place object contains details about that place, such as its name, 
         address and phone number. Extract the name, address, phone number, 
         place ID and place types.
          */
         final CharSequence name = place.getName();
         final CharSequence address = place.getAddress();
         final CharSequence phone = place.getPhoneNumber();
         final String placeId = place.getId();
         String attribution = PlacePicker.getAttributions(data);
         if(attribution == null){
            attribution = "";
         }

         // Update data on map
         mMap.addMarker(new MarkerOptions()
            .position(place.getLatLng())
            .title(name.toString())
           );

         // Print data to debug log
         Log.d(TAG, "Place selected: " + placeId + " (" + name.toString() + ")");


       }
   } else {
       super.onActivityResult(requestCode, resultCode, data);
   }
   // END_INCLUDE(activity_result)
}

After the marker is added to the map, we can easily trigger turn-by-turn navigation by selecting the navigation icon in the bottom-right of the map to launch directions in Google Maps mobile.

Save the Location to the Cloud

Checking in to the places we’ve visited becomes even more useful if we are able to store those check-ins for future reference. Since we’ve already connected our app to the cloud using Firebase, it will be very easy to use that same connection to store our check-ins. After adding the marker to the map, we simply add the following lines of code to our function above to also save the location and id of the place to Firebase.

Map mLocation = new HashMap();
mLocation.put("timestamp", mLastUpdateTime);

Map  mCoordinate = new HashMap();
mCoordinate.put(“latitude”, place.getLatLng().latitude);
mCoordinate.put(“longitude”, place.getLatLng().longitude);
mLocation.put("location", mCoordinate); 

mLocation.put("place_id", place.getId());
myFirebaseRef.push().setValue(mLocation);

That’s all there is to it! We now have an app that, in addition to storing our location history in the cloud, also lets us record a list of all the places we’ve visited while on the road. Stay tuned for future blog posts where we will continue to extend our app’s functionality with features such as directions and support for wearable devices.

Meet Marie Forleo. Named by Oprah as a thought leader for the next generation and one of Inc.’s 500 fastest growing companies of 2014, Marie’s mission is to help people realize their greatest potential and their gifts to change the world. She’s the creator of the award-winning show MarieTV, founder of B-School and has been featured in Inc. Magazine and Entrepreneur Magazine among others. After reading one of her tweets about Google Docs, we chatted to better understand how she uses the family of products to run her business. 

We want to know how you use the Google Docs family of products, too, so share your own examples at +GoogleDocs or @googledocs with the hashtag #mygoogledocs. -Ed. 

Hi Marie. First, tell us about who you are and what you do. 
I’m an unshakable optimist. I believe in a world where the vast majority of people wake up everyday excited to use their gifts to create a kinder, more meaningful life for all. My job is to help you reach your highest potential and use your unique talents to change the world. I do that through writing, teaching, and speaking each week on our show MarieTV, and through our online courses and programs.

How do you guys use Google Docs? 
As a virtual company, our team spans from Hawaii to Los Angeles to New York and more cities in between. And 99.9% of everything we create happens first in Google Docs. We constantly collaborate online and it’s all powered through Google Docs. We write our show scripts, craft and edit newsletters, generate all of the ideas for our free and premium content, brainstorm and execute strategies, track our priorities and goals, and review customer feedback. I’m not kidding when I say Google Docs powers our entire company. We love it and could simply not function at the level we do without it. We have a running joke that if anyone has an idea or initiative they want to bring to life, the first step is “Start a Google doc about it!”

What are three suggestions you’d have for other folks who are using or thinking about using Google Docs to run their business? 

1. Make sure everyone on your team understands Google Docs and is comfortable with how the basics work (sharing, formatting, adding and replying to comments, etc.). A few minutes spent training each other prevents frustration early on and leads to massive gains in productivity down the line. 
2. Be aware of who your documents are shared with. Realize everyone can see comments so be conscious of how you frame feedback to avoid upsets and misunderstandings. 
3. Hot Tip: If you’re collaborating in real time in a super long document, you can quickly and easily “hop to” where a fellow collaborator is writing within the document by clicking on their icon in the upper right corner. It will bring you to exactly where their cursor is. This trick saves us a lot of time and helps us (literally) stay on the same page as each other while working in a virtual environment. 

Do you use Google Docs for anything in your personal life or around the house? 
Oh my goodness, yes. I use Google Docs for everything from tracking family expenses, mapping out home renovation strategies, brainstorming vacation plans and working on personal visions and goals. I seriously LIVE in Google Docs and I love it.

Posted by Alexander Mordvintsev, Software Engineer, Christopher Olah, Software Engineering Intern and Mike Tyka, Software Engineer

Update – 13/07/2015
Images in this blog post are licensed by Google Inc. under a Creative Commons Attribution 4.0 International License. However, images based on places by MIT Computer Science and AI Laboratory require additional permissions from MIT for use.

Artificial Neural Networks have spurred remarkable recent progress in image classification and speech recognition. But even though these are very useful tools based on well-known mathematical methods, we actually understand surprisingly little of why certain models work and others don’t. So let’s take a look at some simple techniques for peeking inside these networks.

We train an artificial neural network by showing it millions of training examples and gradually adjusting the network parameters until it gives the classifications we want. The network typically consists of 10-30 stacked layers of artificial neurons. Each image is fed into the input layer, which then talks to the next layer, until eventually the “output” layer is reached. The network’s “answer” comes from this final output layer.

One of the challenges of neural networks is understanding what exactly goes on at each layer. We know that after training, each layer progressively extracts higher and higher-level features of the image, until the final layer essentially makes a decision on what the image shows. For example, the first layer maybe looks for edges or corners. Intermediate layers interpret the basic features to look for overall shapes or components, like a door or a leaf. The final few layers assemble those into complete interpretations—these neurons activate in response to very complex things such as entire buildings or trees.

One way to visualize what goes on is to turn the network upside down and ask it to enhance an input image in such a way as to elicit a particular interpretation. Say you want to know what sort of image would result in “Banana.” Start with an image full of random noise, then gradually tweak the image towards what the neural net considers a banana (see related work in [1], [2], [3], [4]). By itself, that doesn’t work very well, but it does if we impose a prior constraint that the image should have similar statistics to natural images, such as neighboring pixels needing to be correlated.

So here’s one surprise: neural networks that were trained to discriminate between different kinds of images have quite a bit of the information needed to generate images too. Check out some more examples across different classes:

Why is this important? Well, we train networks by simply showing them many examples of what we want them to learn, hoping they extract the essence of the matter at hand (e.g., a fork needs a handle and 2-4 tines), and learn to ignore what doesn’t matter (a fork can be any shape, size, color or orientation). But how do you check that the network has correctly learned the right features? It can help to visualize the network’s representation of a fork.

Indeed, in some cases, this reveals that the neural net isn’t quite looking for the thing we thought it was. For example, here’s what one neural net we designed thought dumbbells looked like:

There are dumbbells in there alright, but it seems no picture of a dumbbell is complete without a muscular weightlifter there to lift them. In this case, the network failed to completely distill the essence of a dumbbell. Maybe it’s never been shown a dumbbell without an arm holding it. Visualization can help us correct these kinds of training mishaps.

Instead of exactly prescribing which feature we want the network to amplify, we can also let the network make that decision. In this case we simply feed the network an arbitrary image or photo and let the network analyze the picture. We then pick a layer and ask the network to enhance whatever it detected. Each layer of the network deals with features at a different level of abstraction, so the complexity of features we generate depends on which layer we choose to enhance. For example, lower layers tend to produce strokes or simple ornament-like patterns, because those layers are sensitive to basic features such as edges and their orientations.

Left: Original photo by Zachi Evenor. Right: processed by Günther Noack, Software Engineer

Left: Original painting by Georges Seurat. Right: processed images by Matthew McNaughton, Software Engineer

If we choose higher-level layers, which identify more sophisticated features in images, complex features or even whole objects tend to emerge. Again, we just start with an existing image and give it to our neural net. We ask the network: “Whatever you see there, I want more of it!” This creates a feedback loop: if a cloud looks a little bit like a bird, the network will make it look more like a bird. This in turn will make the network recognize the bird even more strongly on the next pass and so forth, until a highly detailed bird appears, seemingly out of nowhere.

The results are intriguing—even a relatively simple neural network can be used to over-interpret an image, just like as children we enjoyed watching clouds and interpreting the random shapes. This network was trained mostly on images of animals, so naturally it tends to interpret shapes as animals. But because the data is stored at such a high abstraction, the results are an interesting remix of these learned features.

Of course, we can do more than cloud watching with this technique. We can apply it to any kind of image. The results vary quite a bit with the kind of image, because the features that are entered bias the network towards certain interpretations. For example, horizon lines tend to get filled with towers and pagodas. Rocks and trees turn into buildings. Birds and insects appear in images of leaves.

The original image influences what kind of objects form in the processed image.

This technique gives us a qualitative sense of the level of abstraction that a particular layer has achieved in its understanding of images. We call this technique “Inceptionism” in reference to the neural net architecture used. See our Inceptionism gallery for more pairs of images and their processed results, plus some cool video animations.

We must go deeper: Iterations

If we apply the algorithm iteratively on its own outputs and apply some zooming after each iteration, we get an endless stream of new impressions, exploring the set of things the network knows about. We can even start this process from a random-noise image, so that the result becomes purely the result of the neural network, as seen in the following images:

Neural net “dreams”— generated purely from random noise, using a network trained on places by MIT Computer Science and AI Laboratory. See our Inceptionism gallery for hi-res versions of the images above and more (Images marked “Places205-GoogLeNet” were made using this network).

The techniques presented here help us understand and visualize how neural networks are able to carry out difficult classification tasks, improve network architecture, and check what the network has learned during training. It also makes us wonder whether neural networks could become a tool for artists—a new way to remix visual concepts—or perhaps even shed a little light on the roots of the creative process in general.