We propose an architecture for fine-grained visual categorization that approaches expert human performance in the classification of bird species. Our architecture first computes an estimate of the object’s pose; this is used to compute local image features which are, in turn, used for classification. The features are computed by applying deep convolutional nets to image patches that are located and normalized by the pose. We perform an empirical study of a number of pose normalization schemes, including an investigation of higher order geometric warping functions. We propose a novel graph-based clustering algorithm for learning a compact pose normalization space. We perform a detailed investigation of stateof-the-art deep convolutional feature implementations [17, 22, 26, 28] and finetuning feature learning for fine-grained classification. We observe that a model that integrates lower-level feature layers with pose-normalized extraction routines and higher-level feature layers with unaligned image features works best. Our experiments advance state-of-the-art performance on bird species recognition, with a large improvement of correct classification rates over previous methods (75% vs. 55-65%). Read More
Monthly Archives: April 2019
Visualizing the AI Revolution in One Infographic
AI Knowledge Map: How To Classify AI Technologies
Is Google AI Aggressive or Cooperative? Depends on the Circumstance
Placing humans at the centre of Artificial Intelligence
At birth, we humans are helpless. We spend about a year unable to walk, about two more before we can articulate full thoughts, and many more years unable to fend for ourselves. We are totally dependent on those around us for our survival. Now compare this to many other mammals. Dolphins, for instance, are born swimming; giraffes learn to stand within hours; a baby zebra can run within forty-five minutes of birth.
Across the animal kingdom, our cousins are strikingly independent soon after they’re born. On the face of it, that seems like a great advantage for other species – but in fact it signifies a limitation. Baby animals develop quickly because their brains are wiring up according to a largely pre-programmed routine. But that preparedness trades off with flexibility.
Now imagine a technology like Artificial Intelligence (AI) that uses an associative data index that shapes itself by the connections that exists in the data. Instead of arriving with everything wired up by a developer for the pre-canned business questions, it knows the connections in the data and allows users to explore the data from any directions and perspectives based on their intuition. This would provide companies with huge flexibility and advantage because every day they have a new business question, and with the “livewired” data, they can explore it and gain unexpected insights. Read More
What’s Your Data Strategy? — Offensive vs Defensive
More than ever, the ability to manage torrents of data is critical to a company’s success. But even with the emergence of data-management functions and chief data officers (CDOs), most companies remain badly behind the curve. Cross-industry studies show that on average, less than half of an organization’s structured data is actively used in making decisions—and less than 1% of its unstructured data is analyzed or used at all. More than 70% of employees have access to data they should not, and 80% of analysts’ time is spent simply discovering and preparing data. Data breaches are common, rogue data sets propagate in silos, and companies’ data technology often isn’t up to the demands put on it.
Having a CDO and a data-management function is a start, but neither can be fully effective in the absence of a coherent strategy for organizing, governing, analyzing, and deploying an organization’s information assets. Indeed, without such strategic management many companies struggle to protect and leverage their data—and CDOs’ tenures are often difficult and short (just 2.4 years on average, according to Gartner). In this article we describe a new framework for building a robust data strategy that can be applied across industries and levels of data maturity. The framework draws on our implementation experience at the global insurer AIG (where DalleMule is the CDO) and our study of half a dozen other large companies where its elements have been applied. The strategy enables superior data management and analytics—essential capabilities that support managerial decision making and ultimately enhance financial performance.
The “plumbing” aspects of data management may not be as sexy as the predictive models and colorful dashboards they produce, but they’re vital to high performance. As such, they’re not just the concern of the CIO and the CDO; ensuring smart data management is the responsibility of all C-suite executives, starting with the CEO. Read More
The new digital divide is between people who opt out of algorithms and people who don’t
Every aspect of life can be guided by artificial intelligence algorithms – from choosing what route to take for your morning commute, to deciding whom to take on a date, to complex legal and judicial matters such as predictive policing.
Big tech companies like Google and Facebook use AI to obtain insights on their gargantuan trove of detailed customer data. This allows them monetize users’ collective preferences through practices such as micro-targeting, a strategy used by advertisers to narrowly target specific sets of users.
In parallel, many people now trust platforms and algorithms more than their own governments and civic society. An October 2018 study suggested that people demonstrate “algorithm appreciation,” to the extent that they would rely on advice more when they think it is from an algorithm than from a human. Read More
The 50 shades of emotion AI
Artificial intelligence offers us an opportunity to amplify service and the integration of technology in everyday lives many times over. But until very recently, there remained a significant barrier in how sophisticated the technology could be. Without a complete understanding of emotion in voice and how AI can capture and measure it, inanimate assistants (voice assistants, smart cars, robots and all AI with speech recognition capabilities) would continue to lack key components of a personality. This barrier makes it difficult for an AI assistant to fully understand and engage with a human operator the same way a human assistant would.
This is starting to change. Rapid advances in technology are enabling engineers to program these voice assistants with a better understanding of the emotions in someone’s voice and the behaviors associated with those emotions. The better we understand these nuances, the more agile and emotionally intelligent our AI systems will become. Read More
Discriminating Systems — Gender, Race, and Power in AI
There is a diversity crisis in the AI sector across gender and race.
The AI sector needs a profound shift in how it addresses the current diversity crisis.
The overwhelming focus on ‘women in tech’ is too narrow and likely to privilege white women over others.
Fixing the ‘pipeline’ won’t fix AI’s diversity problems.
The use of AI systems for the classification, detection, and prediction of race and gender is in urgent need of re-evaluation.
The diversity problem is not just about women. It’s about gender, race, and most fundamentally, about power.10 It affects how AI companies work, what products get built, who they are designed to serve, and who benefits from their development. Read More
The race is on: assessing the US-China Artificial Intelligence competition
Discussions of artificial intelligence are everywhere. Understandably so: AI has a seemingly limitless range applications, from schools to the battlefield. McKinsey & Company estimated that AI is likely to result in $13 trillion of additional global economic activity by 2030. AI also allows the development of autonomous weapons and novel platforms, such as advanced drone swarms. A revanchist Russia might be the scourge of the Western defense community, but Vladimir Putin has arguably issued the clearest articulation of AI’s massive potential: “Whoever becomes the leader in [AI] will become the ruler of the world.” But how do we assess who is leading?
A simple metaphor proves a powerful tool for thinking about the AI race: the traditional manufacturing process. Applying the analogy to the United States vs. China AI competition illustrates that although the United States is ahead overall, China is positioned to surpass it in the long term. On some measures, China is already winning.
A simple manufacturing process consists of three elements: raw materials, production, and manufactured goods. Raw materials are inputs such as wood, wool, or steel. Production includes the equipment, techniques, and manpower to process the raw materials. Manufactured goods are the final outputs: chairs, guns, and tanks. Likewise, current AI systems typically take large amounts of input data, process it using machine-learning techniques, and output trained algorithms. For example, numerous photos of cars can be processed using machine learning to create an algorithm that recognizes cars in other photos. The government and private sector use those algorithms in applications from autonomous vehicle vision to detecting terrorist activity. (Note: machine learning drives the current AI focus, but technically it is only one form of AI. Other forms are less reliant on data.) Read More
Rick's Cafe AI 