Blind Tasting Read online

  Blind Tasting

  by

  A.C. Houston

  Grammarsmith 3rd Edition published 2010

  Copyright © 2010 by A.C. Houston

  Cover design by Duane Clark, Gulf Coast Freelance

  This is a work of fiction. Names, characters, places, brands, media and incidents are either the product of the author’s imagination or are used fictitiously. The author acknowledges the trademarked status and trademark owners of various products referenced in this work of fiction, which have been used without permission. The publication/use of these trademarks is not authorized, associated with, or sponsored by the trademark owners.

  The author expresses deep gratitude to Paul Martin for his role in the original inspiration for this story, explored one summer evening over an excellent bottle of California zinfandel.

  For Teddy

  Chapter One

  VisualAxioms. "So, here's my cost function for estimating region borders in a scene. Recall Duncan's hack using the plenoptic illumination function in face recognition algorithms."

  Shafts of California sunlight fall across Cory Wilder's jeans and black T-shirt as his energetic hand continues to fill a wall-sized whiteboard with mathematical expressions. The light is filtering in through the branches of a eucalyptus tree outside the window.

  John Shang, on the far side of the conference table, leans back in his chair and narrows his eyes analytically at the whiteboard. "We've got 2-D images with occlusions. Optimizing the segmentation of those can, of course, be NP-complete." His low, cultivated voice has a trace of accent from a boyhood lived in Hong Kong.

  Cory pivots to face him. "Right. So iteration ends when there's no significant increase in energy above this adaptive threshold." He taps his cost function with a blue marker. "We stay in polynomial time."

  "Very chill." Pradip Singh beams in satisfaction. He has been following Cory's line of reasoning intently as he stands braced against the edge of the conference table. He's wearing a green VisualAxioms T-shirt that looks vibrant against his dark skin.

  Cory's intense expression dissolves into a smile. "Yeah, I'm hoping you'll come up with the right way to break this into parallel processes."

  Pradip, a native of Bangalore, raises his hands to his face in contemplation, then offers, "I'm thinking compute each camera angle per scene in parallel. The new GPUs are really fast." His speech has the mellifluous cadence of subcontinent syllable-timed English, and the softer execution of 'p' which approximates 'b' to an American ear.

  "So, how's everything going?"

  The three hackers look abruptly toward the doorway at Trish Daniels. They politely conceal their annoyance at this interruption and Trish, who doesn't register it, enters the room with a breezy confidence. At twenty-seven, she's a year older than Cory and John, and two years older than Pradip. She's the new marketing hire and projects a preppy air in polo shirt and slacks, expensive black flats.

  "We've got a new media group, Crowdr, coming in Tuesday, Their backers are Holfield and Tompkins, by the way," she tells them with a touch of awe in her voice. Not getting the enthusiastic reaction she expected, she adds, "Richard thinks there could be some real synergy here."

  "And?" Cory is now leaning against the whiteboard with his arms folded.

  "And they'll want to see what we've got." She now peruses the formulas on the whiteboard and frowns. "I doubt they're going to connect with this Greek, though."

  Cory shakes his head. "Why is Richard pitching this now?"

  She takes a step closer to them, tossing her short blond hair back from her face. "A hundred billion images are captured and uploaded every year onto the Internet and that number just keeps growing. New media is very visual. And, we can provide them with the best visual search technology in the world, right?"

  "Yeah." Cory looks straight at her.

  Trish notes his wide-set brown eyes and his fit, agile body, and thinks he's the hottest geek she's ever worked with. Maintaining her professional tempo, she continues, "So what can I show these guys?"

  Now it's John who answers. He's known as a mathematician who doesn't suffer fools gladly, but his voice is patient at the moment.

  "Our algorithms are still evolving, Trish. This underlying logic is like the rockets, the boosters you need to reach the moon. Your job right now should be to help them understand this process, why we spend so much time on logic. It's critical. Fast, accurate scene analysis by a computer is a hard problem. If it weren't, the solution would be out there already."

  She directs an Ivy League smile at John, but her hazel eyes remain steely. "And, if we run out of money before we get to the moon?" She registers the stress in their faces and presses on. "What is going to get these people excited? The mumbo-jumbo math won't. Cory, we need some sizzle here."

  Cory breathes out and runs his hands through his thatch of brown hair. He looks at her, then points at the whiteboard with a quick, decisive gesture.

  "Guess what? These derivatives and integrals provide a really novel way for a computer to accurately recognize a picture of, say, the Eiffel Tower. Regardless of whether it was taken at night, in the morning, in color, in black and white, from close up, from a panoramic distance, or with varying resolutions of the image." He arches his brows at her. "And, it's going to be an order of magnitude faster and more accurate than anything out there."

  "Can't you make a demo of that?" Trish's expression is energized and commanding.

  Cory nods emphatically. "We will. But we're still working out how to address border detection, rotational deformations, occlusions of objects."

  He's losing her. He moves his hand across the formulas on the whiteboard, rephrasing himself. "These equations are concise statements of how we define visual objects to a computer, so the computer can recognize these objects under a broad set of parameters. The equations guide us around intractabilities. They guide us toward faster, more efficient ways of solving these recognition problems. We have to build and prove this foundation before we can show off any meaningful visual app. The eye candy."

  Eye candy! She crosses her arms, bristling over his choice of words. Is he patronizing her?

  Cory sees her expression and sighs. He points to the last formula he'd written before she came in. "You studied cost functions in B-school, right?" She looks at him incredulously and he can't resist a dig. "So, cool. They do teach calculus, not just the mumbo-jumbo math."

  Her eyes flash at him. Of course she studied cost functions and calculus in her MBA program. But what on earth do these triple integrals on the whiteboard refer to?

  Sharply, she asks, "Have you guys ever considered working with flow charts?"

  When they say nothing, she offers more assertively, "Help me out here, maybe I put this in my PowerPoint presentation. Something like: VisualAxiom's proprietary cost function informs our innovative scene analysis technology. How's that?"

  Cory and John exchange a private look of futility. Pradip stares at the floor.

  She waits for their reply, which finally comes from Pradip.

  "Maybe Trish could use some data visualizations of different object recognition methods. You know, so she could see the measurable trade-offs, the pitfalls, of different approaches. Why our approach is pretty novel. And why it will, hopefully, scale."

  She beams at him. "Yes! Can you put that together in time for Tuesday?"

  Pradip nods. "Probably come in over the weekend and hack up something."

  Cory now pictures his weekend devolving into tedium. But, it's not fair for Pradip to get stuck with the whole thing. He says quietly, "I'll come in Sunday and help you write it."

  Trish notes the word Zinfidel across the front of Cory's T-shirt and recalls hearing that he is into wine. Zinfandel, apparently. Maybe she'll su
ggest a wine bar after work sometime.

  She glances back at John and Pradip. "Okay! We've got a plan. And, remember that Richard wants us all in Sausalito at ten tomorrow. Jo is sending out email with directions to the marina."

  She gives Cory a wry, flirty smile. "Work hard, play hard."

  Chapter Two

  An hour later, Cory returns to his small office. He walks to a shelf containing several books on computer vision, numerical analysis and graph theory and pulls down a volume by Duncan Wright and locates a page. He reviews it quickly and puts the book back, then goes to the whiteboard on the opposite wall and makes some small changes to two mathematical functions written in blue ink. He erases a portion of a diagram on the whiteboard and redraws it, then goes to his desk and rapidly scans email on the thirty-inch flat monitor that's connected to a Mac Pro. Lying underneath the desk is Snoots, who is watching him with an alert, wolfy expression.

  There's nothing in his email that can't wait. He, John and Pradip now have a concise formulation of their new approximation algorithm. They could implement and test it this weekend if it weren't for the visual-search-for-dummies app that Trish now needs asap.

  Cory accepts that Trish is a minefield. From the first day she joined VisualAxioms three weeks ago she has simultaneously tried to dominate him professionally and come on to him in covert ways. He's not interested in her achieving either goal. He isn't going to dwell on that right now, it's Friday evening and he's got plans. He looks at Snoots. "Let's go, boy!"

  The dog crawls out from under the desk, stretches, and looks expectantly at him. Snoots is a three-year-old, seventy-five pound male, a classic black-and-tan shepherd, except for the tips of his ears which flop over slightly. The other trace of yellow lab in his lineage is the lighter coloring around his face, which accentuates his keenly intelligent dark eyes.

  Cory picks up his iPod from the desk, looks at it momentarily, then stuffs it into the back pocket of his jeans. No tunes for the ride home. He pulls on a red hoodie and grabs his book bag off the floor, slinging it onto his shoulders. He moves a ten-speed bicycle away from the wall, removes the black helmet dangling from the handlebars and straps it on his head.

  It's the middle of March and still light when he exits the building. Snoots inspects a nearby shrub, marking it briefly. Cory mounts his bike and heads off down the street with Snoots trotting alongside. They go past a number of small high-tech businesses in this industrial enclave of Menlo Park and Cory notices a new lease sign in front of one space. Someone's entrepreneurial dream didn't pan out.

  He turns the bike up a quiet residential street of modest single-story ranch houses, riding at an easy pace because Snoots is with him. He pats the pocket of his hoodie, confirming the cord leash is there -- a safeguard in case he's confronted for having his dog off-leash. Snoots knows to stay just ahead and to the right of the bicycle when he's on a leash, but it's better exercise for him to run freely and more fun for both of them.

  Cory cycles past colorful flower beds and white cement driveways, reflecting on the past year and how different it would be if Duncan were still running things. Still alive.

  Duncan Wright, a full professor at twenty-eight, left academics to start VisualAxioms to go after a thorny and largely unsolved problem space in computer science -- fast, accurate, dynamic scene recognition. Duncan believed he could scale his innovative concepts into robust technology that would change the game of how visual search and identification are done on the Internet. Or anywhere in the digital universe.

  That was a year ago, about the time Cory completed his Ph.D. thesis at Stanford on parametric geometric constraints for fast boundary detection and Duncan, who served on his committee, wanted Cory to join his startup. The chance to continue working with a scientist of Duncan's stature, and on such a cutting edge venture, was an extraordinary opportunity.

  Cory quit his lucrative part-time gig at a computer game company and joined Duncan's team. Duncan lured John away from CalTech and super-hacker Pradip, who had just completed his masters at Stanford.

  During the first six months the guys basically lived at work. There were late nights together at the whiteboard, dawn runs for caffeinated drinks, bagels and power bars, more whiteboard sessions, intense hacking of ideas to test and measure their algorithms. It was heady, exhilarating.

  A Sand Hill investment firm was backing their effort with seed funding, secured on the basis of Duncan's reputation and the achievement of a couple of preliminary milestones. The investors looked for a business developer so that all the geek power could stay focused on the formidable technical challenges. They found Richard.

  At thirty-six, Richard Dorne is the oldest person at VisualAxioms by almost a decade. His background is digital media. He's neither a scientist nor a hacker, but he successfully grew two startups in the Valley to profitable exits for their investors.

  Cory mostly ignored him in the beginning, assuming Richard could push ahead on the business front with Duncan's reputation and coaching backing him up.

  Duncan was an experienced private pilot, and flew his own Mooney, a high-performance single-engine aircraft which he parked at the Reid-Hillview airport in San Jose. He'd taken Cory, John and Pradip flying a number of times, explaining the instrument panel to them, describing the routine pre-flight check as what you do to convince yourself that this plane can fly now.

  It was a rush of freedom to feel the lift beneath the wings as the runway receded below, to see the tawny foothills looming ahead and experience the angular momentum as Duncan made a smooth turn to head out over the Pacific and turn again to follow the coastline. He often took them over San Gregorio, then north to Half Moon Bay, sometimes flying the plane only a few hundred feet above the water so they could really experience the details of the beaches and farmland from the air. More than one inspired idea about object normalization and edge-detection had been conceived in that rarefied intellectual environment aloft.

  Then, Duncan took his new girlfriend for a sight-seeing trip down the coast. On final approach at a small Santa Cruz county airport, he was forced to make an extreme, evasive maneuver to avoid colliding with another small aircraft that was violating the traffic pattern. Duncan couldn't recover in time to land on the runway and crashed to the side of it. Both he and his girlfriend were killed.

  The other plane was unharmed, but the investigation revealed that its pilot had an outdated chart of the airspace and did not have a functioning radio.

  It's painful to recall this, so Cory focuses on the setting around him. He waves to a familiar elderly man who is watering rose bushes in his front yard. They've never spoken, but the man reminds him a little of his own grandfather back in Ohio. The man waves back, giving Snoots an approving nod.

  The small ranch houses abruptly disappear and are replaced by tall privacy walls with towering green foliage overhanging them. This is Atherton they have just entered. He always enjoys the ride through this peaceful, affluent stretch on the way to the other side of Menlo Park, what used to be unincorporated Menlo Park, where he now lives.

  These stately Atherton manors belong to the entrepreneurial winners of the high-stakes technology pursuits that power this region of the world, Silicon Valley. Or, Silicon Gulch, as it was referred to by the early geeks, decades ago, before the rest of the world knew what was going on out here. Deals move quickly here; a lot of ideas and plans get launched, most disappear, some survive, and a few spawn innovations that create enormous wealth and opportunity for their founders and followers.

  They definitely had a shot at it with Duncan. Do they still?

  For weeks after his death, it was impossible to do any real work. Cory tried to find solace in the abstract beauty of designing algorithms and writing code, but Duncan's absence hung over all of them in a huge, hollow silence. With Duncan, they put their ideas out there and he pushed back hard, demanding more rigor, greater clarity, forcing them to think at deeper levels about the problem.

  Richard persuaded the inve
stors that he could keep the company on track, at least for the near term. He emphasized the ability of the team Duncan had put in place with Cory, John and Pradip, and drafted a rough schedule of new milestones which he presented to them.

  But, they are now a ship without a captain. They need an advocate who understands their work they way they understand it, someone who can communicate it to the world of investors. Why did Richard expend precious resources to hire someone like Trish? They are burning through their capital and Cory knows that getting another round of funding is far less certain than it was six months ago with Duncan at the helm.

  They're past Atherton now, coming up to El Camino Real, a bustling thoroughfare. Cory watches Snoots; the dog knows to wait. They move across the intersection and head up a less busy street, residential again, back to the small ranch houses. Cory wonders if Becca wants to move in with him.

  He recalls the day he met her at Cafe Boronne a few weeks after Duncan's plane crash. The patio tables were packed with people and he'd been sitting there for hours in a fugue-like concentration, implementing a feature-extraction algorithm on his laptop. A soft, feminine voice asked to share his table and he looked up to see an extremely pretty girl in a jean jacket holding a cappuccino and pastry, her light brown hair falling past her waist.

  He said sure, a little caught off guard, but with growing admiration of her confident, airy manner. He shifted from writing code to having a conversation with her.

  He learned she had just finished her undergrad degree in communications at Stanford and was working for a graphics design firm in Redwood City. She knew someone at the game company he used to work for, too.

  Cory gave her a broad overview of what he did and, that day, she seemed genuinely interested. He usually felt at ease in any intellectual discussion, but Becca's startling beauty was unsettling. Nevertheless, he made sure to exchange contact information.