RL’s origins and historic context
RL copies a very simple principle from nature. The psychologist Edward Thorndike documented it more than 100 years ago. Thorndike placed cats inside boxes from which they could escape only by pressing a lever. After a considerable amount of pacing around and meowing, the animals would eventually step on the lever by chance. After they learned to associate this behaviour with the desired outcome, they eventually escaped with increasing speed.
Some of earliest AI researchers believed that this process might be usefully reproduced in machines. In 1951, Marvin Minsky, a student at Harvard who would become one of the founding fathers of AI, built a machine that used a simple form of reinforcement learning to mimic a rat learning to navigate a maze. Minsky’s Stochastic Neural Analogy Reinforcement Computer (SNARC), consisted of dozens of tubes, motors, and clutches that simulated the behaviour of 40 neurons and synapses. As a simulated rat made its way out of a virtual maze, the strength of some synaptic connections would increase, thereby reinforcing the underlying behaviour.
There were few successes over the next few decades. In 1992, Gerald Tesauro demonstrated a program that used the technique to play backgammon. It became skilled enough to rival the best human players, a landmark achievement in AI. But RL proved difficult to scale to more complex problems.
In March 2016, however, AlphaGo, a program trained using RL, won against one of the best Go players of all time, South Korea’s Lee Sedol. This milestone event opened again teh pandora’s box of research about RL. Turns out the key to having a strong RL is to combine it with deep learning.
Current usage and major methods of RL
Thanks to current RL research, computers can now automatically learn to play ATARI games, are beating world champions at Go, simulated quadrupeds are learning to run and leap, and robots learn how to perform complex manipulation tasks that defy explicit programming.
However, while RL saw its advancements accelerate, progress in RL has not been driven as much by new ideas or additional research as just by more of data, processing power and infrastructure. In general, there are four separate factors that hold back AI:
- Processing power (the obvious one: Moore’s Law, GPUs, ASICs),
- Data (in a specific form, not just somewhere on the internet – e.g. ImageNet),
- Algorithms (research and ideas, e.g. backprop, CNN, LSTM), and
- Infrastructure (Linux, TCP/IP, Git, AWS, TensorFlow,..).
Similarly for RL, for example for computer vision, the 2012 AlexNet (deeper and wider version of 1990’s Convolutional Neural Networks – CNNs). Or, ATARI’s Deep Q Learning is an implementation of a standard Q Learning algorithm with function approximation, where the function approximator is a CNN. AlphaGo uses Policy Gradients with Monte Carlo tree search (MCTS).
RL’s most optimal method vs. human learning
Generally, RL approaches can be divided into two core categories. The first focuses on finding the optimum mappings that perform well in the problem of interest. Genetic algorithm, genetic programming and simulated annealing have been commonly employed in this class of RL approaches. The second category is to estimate the utility function of taking an action for the given problem via statistical techniques or dynamic programming methods, such as TD(λ) and Q-learning. To date, RL has been successfully applied in many real-world complex applications, including autonomous helicopter, humanoid robotics, autonomous vehicles, etc.
Policy Gradients (PGs), one of RL’s most used methods, is shown to work better than Q Learning when tuned well. PG is preferred because there’s an explicit policy and a principled approach that directly optimises the expected reward.
Before trying PGs (canon), it is recommended to first try to use cross-entropy method (CEM) (normal gun), a simple stochastic hill-climbing “guess and check” approach inspired loosely by evolution. And if you really need to or insist on using PGs for your problem, use a variation called TRPO, which usually works better and more consistently than vanilla PG in practice. The main idea is to avoid parameter updates that change the policy dramatically, as enforced by a constraint on the KL divergence between the distributions predicted by old and the new policies on data.
PGs, however have few disadvantages: they typically converge to a local rather than a global optimum and they display inefficient and high variance while evaluating a policy. PGs also require lot of training samples, take lot of time to train, and are hard to debug debug when they don’t work.
PG is a fancy form of guess-and-check, where the “guess” refers to sampling rollouts from a current policy and encouraging actions that lead to good outcomes. This represents the state of the art in how we currently approach RL problems. But compare that to how a human might learn (e.g. a game of Pong). You show him/her the game and say something along the lines of “You’re in control of a paddle and you can move it up or down, and your goal is to bounce the ball past the other player”, and you’re set and ready to go. Notice some of the differences:
- Humans communicate the task/goal in a language (e.g. English), but in a standard RL case, you assume an arbitrary reward function that you have to discover through environment interactions. It can be argued that if a human went into a game without knowing anything about the reward function, the human would have a lot of difficulty learning what to do but PGs would be indifferent, and likely work much better.
- A human brings in a huge amount of prior knowledge, such as elementary physics (concepts of gravity, constant velocity,..), and intuitive psychology. He/she also understands the concept of being “in control” of a paddle, and that it responds to your UP/DOWN key commands. In contrast, algorithms start from scratch which is simultaneously impressive (because it works) and depressing (because we lack concrete ideas for how not to).
- PGs are a brute force solution, where the correct actions are eventually discovered and internalised into a policy. Humans build a rich, abstract model and plan within it.
- PGs have to actually experience a positive reward, and experience it very often in order to eventually shift the policy parameters towards repeating moves that give high rewards. On the other hand, humans can figure out what is likely to give rewards without ever actually experiencing the rewarding or unrewarding transition.
In games/situations with frequent reward signals that requires precise play, fast reflexes, and not much planning, PGs quite easily can beat humans. So once we understand the “trick” by which these algorithms work you can reason through their strengths and weaknesses.
PGs don’t easily scale to settings where huge amounts of exploration are difficult to obtain. Instead of requiring samples from a stochastic policy and encouraging the ones that get higher scores, deterministic policy gradients use a deterministic policy and get the gradient information directly from a second network (called a critic) that models the score function. This approach can in principle be much more efficient in settings with high-dimensional actions where sampling actions provide poor coverage, but so far seems empirically slightly finicky to get working.
There is also a line of work that tries to make the search process less hopeless by adding additional supervision. In many practical cases, for instance, one can obtain expert trajectories from a human. For example AlphaGo first uses supervised learning to predict human moves from expert Go games and the resulting human mimicking policy is later fine-tuned with PGs on the “real” goal of winning the game.
RL’s new frontiers: MAS, PTL, evolution, memetics and eTL
There is another method called Parallel Transfer Learning (PTL), which aims to optimize RL in multi-agent systems (MAS). MAS are computer systems composed of many interacting and autonomous agents within an environment of interests for problem-solving. MAS have a wide array of applications in industrial and scientific fields, such as resource management and computer games.
In MAS, as agents interact with and learn from one another, the challenge is to identify suitable source tasks from multiple agents that will contain mutually useful information to transfer. In conventional MAS (cMAS), which are optimal for simple environments, actions of each agent are pre-defined for possible states in the environment. Normal RL methodologies have been used as the learning processes of (cMAS) agents through trial-and-error interactions in a dynamic environment.
In PTL, each agent will broadcast its knowledge to all other agents while deciding whose knowledge to accept based on the reward received from other agents vs. expected rewards it predicts. Nevertheless, agents in this approach tend to infer incorrect actions on unseen circumstances or complex environments.
However, for more complex or changing environments, it is necessary to endow the agents with intelligence capable of adapting to an environment’s dynamics. A complex environment, almost by definition, implies complex interactions and necessitated learning of MAS, which current RL methodologies are hard-pressed to meet. A more recent machine learning paradigm of Transfer Learning (TL) was introduced as an approach of leveraging valuable knowledge from related and well studied problem domains to enhance problem-solving abilities of MAS in complex environments. Since then, TL has been successfully used for enhancing RL tasks via methodologies such as instance transfer, action-value transfer, feature transfer and advice exchanging (AE).
Most RL systems aim to train a single agent or cMAS. Evolutionary Transfer Learning framework (eTL) aims to develop intelligent and social agents capable of adapting to the dynamic environment of MAS and more efficient problem solving. It’s inspired by Darwin’s theory of evolution (natural selection + random variation) by principles that govern the evolutionary knowledge transfer process. eTL constructs social selection mechanisms that are modelled after the principles of human evolution. It mimics natural learning and errors that are introduced due to the physiological limits of the agents’ ability to perceive differences, thus generating “growth” and “variation” of knowledge that agents have, thus exhibiting higher adaptability capabilities for complex problem solving. Essential backbone of eTL comprises of memetic automaton, which includes evolutionary mechanisms such as meme representation, meme expression, etc.
Memetics in eTL
eTL implementation with learning agents
- MASs with TL vs. MAS without TL: Most TL approaches outperform cMAS. This is due to TL endowing agents with capacities to benefit from the knowledge transferred from the better performing agents, thus accelerating the learning rate of the agents in solving the complex task more efficiently and effectively.
- eTL vs. PTL and other TL approaches: FALCON and BP agents with the eTL outperform PTL and other TL approaches due to the reason that, when deciding whether to accept information broadcasted by the others, agents in PTL tend to make incorrect predictions on previously unseen circumstances. Further, eTL also demonstrates superiority in attaining higher success rates than all AE models thanks to meme selection operator of eTL, which considers a fusion of the “imitate-from-elitist” and “like-attracts-like” principles so as to give agents the option of choosing more reliable teacher agents over the AE model.
While popularisation of RL is traced back to Edward Thorndike and Marvin Minsky, it’s been inspired by nature and present with us humans since ages long gone. This is how we effectively teach children and want to now teach our computer systems, real (neural networks) or simulated (MAS).
RL reentered human consciousness and rekindled our interest again in 2016 when AlphaGo beat Go champion Lee Sedol. RL has, via its currently successful PGs, DQNs and other methodologies, already contributed and continues to accelerate, turn more intelligent and optimise humanoid robotics, autonomous vehicles, hedge funds, and other endeavours, industries and aspect of human life.
However, what is that optimises or accelerates RL itself? Its new frontiers represent PTLs, Memetics and a holistic eTL methodology inspired by natural evolution and spreading of memes. This latter evolutionary (and revolutionary!) approach is governed by several meme-inspired evolutionary operators (implemented using FALCON and BP multi-layer neural network), including meme evolutions.
The performance efficacy of eTL seems to have outperformed even most state-of-the-art MAS TL systems (PTL).
What future does RL hold? We don’t know. But the amount of research resources, experimentation and imaginative thinking will surely not disappoint us.
Nowadays, ‘artificial intelligence’ (AI) and ‘machine learning’ (ML) are cliches that people use to signal awareness about technological trends. Companies tout AI/ML as panaceas to their ills and competitive advantage over their peers. From flower recognition to an algorithm that won against Go champion to big financial institutions, including ETFs of the biggest hedge fund in the world are already or moving to the AI/ML era.
However, as with any new technological breakthroughs, discoveries and inventions, the path is laden with misconceptions, failures, political agendas, etc. Let’s start by an overview of basic methodologies of ML, the foundation of AI.
101 and limitations of AI/ML
The fundamental goal of ML is to generalise beyond specific examples/occurrences of data. ML research focuses on experimental evaluation on actual data for realistic problems. ML’s performance is then evaluated by training a system (algorithm, program) on a set of test examples and measuring its accuracy at predicting the novel test (or real-life) examples.
Most frequently used methods in ML are induction and deduction. Deduction goes from the general to the particular, and induction goes from the particular to the general. Deduction is to induction what probability is to statistics.
Let’s start with induction. Domino effect is perhaps the most famous instance of induction. Inductive reasoning consists in constructing the axioms (hypotheses, theories) from the observation of supposed consequences of these axioms.Induction alone is not that useful: the induction of a model (a general knowledge) is interesting only if you can use it, i.e. if you can apply it to new situations, by going somehow from the general to the particular. This is what scientists do: observing natural phenomena, they postulate the laws of Nature. However, there is a problem with induction. It’s impossible to prove that an inductive statement is correct. At most can one empirically observe that the deductions that can be made from this statement are not in contradiction with experiments. But one can never be sure that no future observation will contradict the statement. Black Swam theory is the most famous illustration of this problem.
Deductive reasoning consists in combining logical statements (axioms, hypothesis, theorem) according to certain agreed upon rules in order to obtain new statements. This is how mathematicians prove theorems from axioms. Proving a theorem is nothing but combining a small set of axioms with certain rules. Of course, this does not mean proving a theorem is a simple task, but it could theoretically be automated.
A problem with deduction is exemplified by Gödel’s theorem, which states that for a rich enough set of axioms, one can produce statements that can be neither proved nor disproved.
Two other kinds of reasoning exist, abduction and analogy, and neither is frequently used in AI/ML, which may explain many of current AI/ML failures/problems.
Like deduction, abduction relies on knowledge expressed through general rules. Like deduction, it goes from the general to the particular, but it does in an unusual manner since it infers causes from consequences. So, from “A implies B” and “B”, A can be inferred. For example, most of a doctor’s work is inferring diseases from symptoms, which is what abduction is about. “I know the general rule which states that flu implies fever. I’m observing fever, so there must be flu.” However, abduction is not able to build new general rules: induction must have been involved at some point to state that “flu implies fever”.
Lastly, analogy goes from the particular to the particular. The most basic form of analogy is based on the assumption that similar situations have similar properties. More complex analogy-based learning schemes, involving several situations and recombinations can also be considered. Many lawyers use analogical reasoning to analyse new problems based on previous cases. Analogy completely bypasses the model construction: instead of going from the particular to the general, and then from to the general to the particular, it goes directly from the particular to the particular.
Let’s next check some of conspicuous failures in AI/ML (in 2016) and corresponding AI/ML methodology that, in my view, was responsible for failure:
Microsoft’s chatbot Tay utters racist, sexist, homophobic slurs (mimicking/analogising failure)
In an attempt to form relationships with younger customers, Microsoft launched an AI-powered chatbot called “Tay.ai” on Twitter in 2016. “Tay,” modelled around a teenage girl, morphed into a “Hitler-loving, feminist-bashing troll“—within just a day of her debut online. Microsoft yanked Tay off the social media platform and announced it planned to make “adjustments” to its algorithm.
AI-judged beauty contest was racist (deduction failure)
In “The First International Beauty Contest Judged by Artificial Intelligence,” a robot panel judged faces, based on “algorithms that can accurately evaluate the criteria linked to perception of human beauty and health.” But by failing to supply the AI/ML with a diverse training set, the contest winners were all white.
Chinese facial recognition study predicted convicts but shows bias (induction/abduction failure)
Researchers in China’s published a study entitled “Automated Inference on Criminality using Face Images.” They “fed the faces of 1,856 people (half of which were convicted violent criminals) into a computer and set about analysing them.” The researchers concluded that there were some discriminating structural features for predicting criminality, such as lip curvature, eye inner corner distance, and the so-called nose-mouth angle. Many in the field questioned the results and the report’s ethics underpinnings.
The above examples must not discourage companies to incorporate AI/ML into their processes and products. Most AI/ML failures seem to stem from band-aid, superfluous way of embracing AI/ML. A better and more sustainable approach to incorporating AI/ML would be to initiate a mix of projects generating both quick-wins and long-term transformational products/services/process. For quick-wins, a company might focus on changing internal employee touchpoints, using recent advances in speech, vision, and language understanding, etc.
For long-term projects, a company might go beyond local/point optimisation, to rethinking business lines, products/services, end-to-end processes, which is the area in which companies are likely to see the greatest impact. Take Google. Google’s initial focus was on incorporating ML into a few of their products (spam detection in Gmail, Google Translate, etc), but now the company is using machine learning to replace entire sets of systems. Further, to increase organisational learning, the company is dispersing ML experts across product groups and training thousands of software engineers, across all Google products, in basic machine learning.