I bet this sounds familiar: “What the heck is that body of water I’m looking at right now? I can’t tell if that’s just a really big lake or an ocean!” It’s a question we have all asked. The realities of that question are far more complex, and often have a far reaching impact on a company’s bottom line.
What is an ocean?
The official definition of an ocean leaves much to be desired. And it’s not going to help you much when you find yourself confounded before a vast aquatic expanse. Perhaps what you see before you is a lake? Or perhaps a sea? Or, most maddening of all, a giant pond? Very different from a lake. There’s simply no way to know. And if you’re lost in the wilderness, exploring the uncharted reaches of our planet, or suddenly and unexpectedly found adrift – knowing an ocean from a pond can be the difference between life and death.
Ocean Recognition can also be a costly expense for businesses who choose to locate themselves near water. We at Xnor enjoy a lovely view of Lake Union just outside our windows, but that view comes with a cost. We had to develop our own proprietary employee onboarding to assist in differentiating Lake Union from other prolific local bodies of water near us such as: Lake Washington, Puget Sound, and the Pacific Ocean. Prior to our training, countless man hours were lost to confused employees staring out the window in bewilderment, or driving in circles around Lake Washington believing themselves on the way to our office. And we know we are not alone in this.
Xnor to the rescue
While many other companies have focused their attention on solving mundane business problems, Xnor is dedicated to helping people – especially those who are continuously circling bodies of water with no hope in sight. Here’s a video of our development process in action creating and iterating on V.11 of our Ocean Recognizer 9000.
The Ocean Recognizer 9000 is a culmination of work by the greatest minds in AI, hardware engineering, lifeguarding, and of course, water science. With such incredible individuals working toward a common goal, we knew we could develop a technology that would put an end to this crisis once and for all.
So How does Ocean Recognition work?
The problem of reliable Ocean Recognition is confounded by the subtlety of the visual cues available in a limited visual scene that does not circumscribe the body of water. Instead, inference must proceed from features that require a deep understanding of the scene. How pronounced is the gravitational pull of the moon? Is the flora and fauna of the scene commensurate with oceanlike salinity? Are visible ships humble vessels or equipped for bona-fide seafaring?
Recent breakthroughs in Deep Neural Networks permit rich representations of complex conceptual features like the above to be learned by a suitably trained model. We gathered rich visual imagery of the world’s oceans spanning continents and environmental conditions for training. Hard negative mining was employed to teach the networks to be discriminative of the world’s largest lakes and bays. Transfer learning and recurrent neural networks allowed leverage progress made in natural language and sequential modeling techniques so that patterns in the majestic rolling waves would not be lost upon a single frame recognizer.
To make up for the small available sample size of the worlds available oceans, Generative Adversarial Networks (GANs) competed in a feedback loop to synthesize realistic positive and negative examples of oceanscapes in lands and seas that never were. By selectively exciting the model, were able to isolate individual neurons in the network associated with high level features such as whale spouts and salt stained driftwood.
Our resulting network is about as deep as the Marianas Trench, so we leverage Xnor’s model compaction and quantization technologies as well as some other proprietary techniques for efficient inference so that when you are standing at the edge, gazing into an abyss, then Xnor’s ocean Recognizer can gaze with you.
How do you get Ocean recognition for your business?
While distribution of the Ocean Recognizer 9000 was initially a direct to consumer effort, we realized that the best path to monetization would be through businesses (not that we’re doing this for money or anything). Initially most businesses were unsure about why they should buy the Ocean Recognizer 9000. They asked questions like: What is this thing? Why does it keep falling apart? Why would I ever need something like this? So instead of traditional business channels, we decided to lobby Congress to require every office in America be equipped with our Ocean Recognizer. It’s senseless to have fire extinguishers in every building to protect us from fire, but nothing to protect us from the hazards of large bodies of water.
If your business is still looking to acquire Ocean Recognizers, please fax a request letter to: 206-602-6283, ATTN: Frank Ocean. We will fulfill your order within 5-6 business months. If you order now, you’ll also receive Xnor’s fully sentient AI Bot (just pay for shipping and handling).