The truth about deep learning

Anyone following DL news knows that DL alone will not lead to strong AI. The most impressive feats in the last year or so have come from combining deep artificial neural networks with other algorithms, just as DeepMind combined deep ConvNets with reinforcement learning and Monte Carlo Tree Search. There&#x27;s not really an interesting conversation to be had about whether DL will get us to strong AI. It won&#x27;t. It is just machine perception; that is, it classifies, clusters and makes predictions about data very well in many situations, but it&#x27;s not going to solve goal-oriented learning. But it solves perception problems very well, often better than human experts. So in the not too distant future, as people wake up to its potential, we will use those infinitely replicable NNs to extract actionable knowledge from the raw data of the world. That is, the world will become more transparent. It will offer fewer surprises. We may not solve cancer with DL, but we will spot it in X-rays more consistently with image recognition, and save more lives.<p>Disclosure: I work on the open-source DL project Deeplearning4j: <a href="http:&#x2F;&#x2F;deeplearning4j.org&#x2F;" rel="nofollow">http:&#x2F;&#x2F;deeplearning4j.org&#x2F;</a>

I understand and empathize with the skepticism or rather criticisms around hand wringing with respect to the implications of current deep learning methods.<p>However, as someone who builds them for vision applications I&#x27;m increasingly convinced that some form of ANN will underlie AGI - what he calls a universal algorithm.<p>If we assume that general intelligence comes from highly trained, highly connected single processors (neurons) with a massive and complex sensor system, then replicating that neuron is step one - which arguably is what we are building, albeit comparatively crudely, with ANN&#x27;s.<p>If you compare at a high level how infants learn and how we train RNN&#x2F;CNNs they are remarkably similar.<p>I think where the author, and in general the ML crowd focuses too much is on unsupervised learning as being pivotal for AGI.<p>In fact if you look again at biological models the bulk of animal learning is supervised training in the strict technical sense. Just look at feral children studies as proof of this.<p>Where the author detours too much is assuming the academic world would prove a broader scope for ANN if it were there. In fact however research priorities are across the board not focused on general intelligence and most machine learning programs explicitly forbid this research for graduate students as it&#x27;s not productive over the timeline of a program.<p>Bengio and others I think are on the right track, focusing on the question of ANN towards AGI and I think it will start producing results as our training methods.

&quot;Here is my personal answer to the second question: deep neural networks are more useful than traditional neural networks for two reasons: The automatic encoding of features which previously had to be hand engineered. The exploitation of structurally&#x2F;spatially associated features. At the risk of sounding bold, that’s it — if you believe there is another benefit which is not somehow encompassed by these two traits, please let me know.&quot;<p>Let me ask a very simple question. What set of hand-engineered features gives &lt;5% error on ImageNet?

My top reasons why everyone getting into maths&#x2F;stats&#x2F;cs should go straight for deep learning:<p>a. recent findings are documented incredibly well in both research and code<p>b. because of its success, there are many areas for useful contribution at relatively less effort from the researcher<p>c. because of its success, it&#x27;ll help you develop marketable skills<p>d. it&#x27;s fun<p>Maybe it won&#x27;t solve General AI, but it seems like a damn good foundation for the person&#x2F;people that will eventually come out with ideas that move us closer in that direction.

&quot;The automatic encoding of features which previously had to be hand engineered.&quot; Yes, that is the main benefit.<p>The drawback is that we are still hand tuning architectures, slowly inventing (or incorporating) things like LSTM and the like into the model.<p>One goal would be to achieve a universal building block that can be stacked&#x2F;repeated without the need for architectural tuning.<p>Maybe something that combines recurrence, one-shot learning, deep learning, and something stolen from AI (like alpha-beta, graph search, or something self-referencing and stochastic with secondary neural networks) into a single &quot;node&quot;. Then we won&#x27;t have to worry about architecture so much.

Why are we so preoccupied with the notion of &quot;artificial intelligence&quot; in the first place?<p>Artificial intelligence, if it can even be defined, does not seem like a particularly valuable goal. Why is emulating human cognition the metric by which we assess the utility of machine learning systems?<p>I&#x27;d take a bunch of systems with superhuman abilities in specialized fields (driving, Go, etc.) over so-called &quot;artificial intelligence&quot; any day.

What are &quot;traditional nets&quot; ? What are the &quot;other learning algorithms&quot; ? What is a universal algorithm (and for what)? Neural nets are universal function approximators. There isnt something [edit: a function] they can&#x27;t learn. When stacked they seem to produce results that are eerily human-like.<p>I think the &quot;universal algorithm&quot; in the article refers to some kind of emergent intelligence. Well, nothing that he mentions precludes it. Our brains aren&#x27;t magical machines. Neural nets may not model real neurons, yet it is amazing how they can produce results that we identify as similar to the way we think. There is nothing in computational neuroscience that comes close to this. If anything, the success of deep nets bolsters my belief in connectionism rather than the opposite. I would expect it is very difficult to formulate &quot;intelligence&quot; mathematically, and to prove that DNs can or cannot produce it.

The truth about most automation in general:<p>The logic is written by humans. The <i>main</i> mechanism by which computers &#x2F; robots begin to outperform people in eg playing Chess, Go or Driving, is <i>copying what works</i>.<p>Humans outperformed animals because they were able to try stuff, recognize what works and transmit that abstract information using language.<p>The main advantage of computers is being able to <i>quickly and easily copy bits</i> and check for errors. You can have perfect copies now, preserving things that before could only be copied imperfectly.<p>And now you copy algorithms that work. The selection process might need work but the actual logic is still written by some human somewhere. It&#x27;s almost never written by a computer. Almost all the code is actually either written by a human or at most generated by an algorithm written by a human, which takes as input code written by another human.<p>What&#x27;s the &quot;smarter&quot; thing is the system of humans banging away at a platform, all making little contributions, and the selection process for what goes into the next version. That&#x27;s what&#x27;s smarter than a single human. That and the ability to collect and process tons of data.<p>All the current AI does is throw a lot of machines at a problem, and stores the result in giant databases as precomputed input for later. That&#x27;s what most <i>big data</i> is today. Whoever has the training sets and the results is now hoarding it for a competitive advantage.<p>But really, the thing that makes all the system smart is that so many humans can make their own diff&#x2F;patch&#x2F;&quot;pull request&quot;. Anyone can write a test and submit a bug that something doesn&#x27;t work. That openness what made science and open source successful.<p>Open source has served the <i>long tail</i> better, too. Microsoft builds software that runs on some hardware. Linux has been forked to run on <i>toasters</i>. Open source drug platforms would have helped solve malaria, zika and other diseases faster.<p>If we had patentleft in drugs, we&#x27;d outpace bacterial resistance. Instead we have the profit motive, which stagnated development of new drugs.

Not sure why this is so highly upvoted. Nobody is questioning that deep networks work better than shallow ones, and there is a good understanding in academia of why (that fits with most lay people&#x27;s intuition). I hardly consider that the most interesting or relevant question.

&quot;Since I am feeling especially bold, I will make another prediction: deep learning will not produce the universal algorithm. There is simply not enough there to create such a complex system.&quot;<p>While I (emotionally) agree, it will be interesting to see if the complexity (and non-linearity) of these algorithms permit &#x27;emergent&#x27; behavior to appear.

From my intuitive understanding (not an expert), very abstract description how it works in general: - you have real world problem -&gt; task which you need to solve - you build model (algorithm, math method &amp; etc) which should solve the task - you need to find optimum of the complex function (error function)<p>Third step is usually finding optimum of the function. Deep neural networks help you to move complexity from step 2 to step 3. One example you mentioned, when feature engineering is moved from 2 -&gt; 3. So you can use simpler methods on step2 to solve same problems, or extend problems area which you can solve with the same complexity on step2.

Can anyone recommend a good resource that summarises what the different algorithms are best suited for aimed at novices?<p>I&#x27;ve been working my way through <a href="http:&#x2F;&#x2F;neuralnetworksanddeeplearning.com&#x2F;" rel="nofollow">http:&#x2F;&#x2F;neuralnetworksanddeeplearning.com&#x2F;</a> (with a big detour back into maths thanks to the Khan Academy) and have done a few ML courses, but they mainly cover a couple of algorithms, not all the ones available in spark&#x27;s MLLib or tensorflow for example.

In my opinion, in the 80-90&#x27;s, neural networks and machine learning used to be 10% a solid concept in terms of academic research and 90% hype. Now neural networks and machine learning are 10% a solid concept in terms of being a practical applicable tool and 90% hype. Things have changed a lot because I almost run out of fingers when trying to express the orders of magnitude in which raw processing power has increased. You can literally feed the network with anything when training and get reasonable results later in recognition. That&#x27;s one impressive yet humanly vague hash table there. And no, you don&#x27;t have to wait for months or weeks anymore to train new things. Not even days, necessarily.<p>Why people pull in artificial intelligence is both naively optimistic and quite understandable. Modelling something of a neural system is so close to how biological brain works that the parallel is blatantly obvious. On the other hand, the current deep networks do not translate to intelligence; not at all. Machine learning might be, in part, something we could describe as &quot;intelligent&quot; as it&#x27;s able to connect dots that are very difficult to connect by traditional algorithms but it absolutely is no intelligence. Then again, we do hang out in the same neighbourhood. If we will ever create an artificial intelligence in software I&#x27;m quite certain it will be very much based on some sort of massively deep and parallel network of dynamic connections.<p>I&#x27;m not that interested in artificial intelligence myself. I would be interested in artificial creativity and emotional senses, but to model those there are bigger metaphysical questions to be answered first.

I love the last sentence, and want to expand on it. If ANNs are tools to help computers perceive, then they are analogous to components or layers in the nervous system. If we map the nervous system thoroughly enough and understand the inputs and outputs of each layer&#x2F;region, then reproducing a human-like nervous system might not be all that complicated.

People have been working on neural nets for over 50 years now. The topic goes in and out of fashion. Nets are more powerful now and computers vastly more powerful. <a href="https:&#x2F;&#x2F;en.m.wikipedia.org&#x2F;wiki&#x2F;Perceptrons_(book)" rel="nofollow">https:&#x2F;&#x2F;en.m.wikipedia.org&#x2F;wiki&#x2F;Perceptrons_(book)</a>

When you have to train a network with a zillion images of a dumbbell (<a href="http:&#x2F;&#x2F;www.businessinsider.sg&#x2F;googles-ai-can-teach-us-about-the-human-brain-2015-7&#x2F;#.V1AWhT9VK1E" rel="nofollow">http:&#x2F;&#x2F;www.businessinsider.sg&#x2F;googles-ai-can-teach-us-about-...</a>) for it to recognise what a dumbbell is and then it still gets it wrong (adding arms!), then somethings fundamentally broken, in so much humans don&#x27;t learn like that. DL is a huge step forward but it&#x27;s not ever going to be any kind of AGI.

As usual with tools, even these, a clear understanding of the specific problem, the relevant metrics and the expected goal is decisive. I am saying that experimental protocols are still devised by humans against a cost vs opportunity matrix. Brute computational force is not independent yet, artificial intelligence has not emerged yet.

This article shows how desp learning is different than true <i>human-like</i> understanding:<p><a href="http:&#x2F;&#x2F;www.wired.com&#x2F;2016&#x2F;03&#x2F;doug-lenat-artificial-intelligence-common-sense-engine&#x2F;" rel="nofollow">http:&#x2F;&#x2F;www.wired.com&#x2F;2016&#x2F;03&#x2F;doug-lenat-artificial-intellige...</a>

&gt; deep learning will not produce the universal algorithm<p>I&#x27;m curious what HN users think the &quot;universal algorithm&quot; will end up looking like?<p>My own guess (wild speculation) is that we&#x27;ll start moving in the direction of concepts like tensor networks. While that term sounds like it has something to do with machine learning, it actually falls under the domain of theoretical physics. Tensor networks are a relatively recent development in quantum mechanics that show promise because of their ability to extract the &quot;interesting&quot; information from a quantum state. Generally speaking, it&#x27;s very difficult to compute&#x2F;describe&#x2F;compress a quantum state because it &quot;lives&quot; in an exponentially large Hilbert space. Traditionally, the field of quantum chemistry has built this space up using Gaussian basis functions, and the field of solid state physics has built it up using plane waves. The problem is that regardless of the basis set chosen, it appears as though exponentially more basis vectors are required to accurately describe a quantum state as the system becomes larger.<p>Tensor networks are an attempt to alleviate this problem. While it is true that the state space of an arbitrary quantum system is exponentially large in the number of particles, it turns out that for <i>realistic</i> quantum systems, the relevant state space is actually much smaller — i.e., real systems seem to live in a tiny corner of Hilbert space. And this tiny subspace even includes all of the possible states that one could put a collection of qubits into within the lifetime of the universe.<p>The projection of a system&#x27;s state vector into either the position or momentum basis is known as the system&#x27;s &quot;wavefunction&quot; (some texts allow more than these two bases). Since the wavefunction exhibits the highly desirable property of being localized in position&#x2F;momentum space, this allows one to build up a good approximation to the state using Gaussians or plane waves — that is, unless the wavefunction exhibits strong electron correlation (quantum entanglement). Quantum entanglement is the exception to nature&#x27;s tendency to localize state space about a point in spacetime, and thus it is frequently the case that the most commonly used basis sets are highly suboptimal for many real electronic systems (superconductors stand out as a notable and somewhat pathological example).<p>I&#x27;m not entirely familiar with all of the math behind it, but tensor networks essentially describe the small but relevant region of Hilbert space by exploiting properties of the renormalization group. In this sense, a compact way of describing &quot;real world&quot; quantum states is developed. I think this has applications to a &quot;universal algorithm&quot;, because real world data rarely consists of a random or uniform scattering of information across the data&#x27;s state space. In my own research, I&#x27;ve found that a lot of the NP-hard problems I run into are efficiently solvable in practice (stuff involving low rank PSD matrices) precisely because the data <i>isn&#x27;t</i> random. If tensor networks are good at finding a basis set that is &quot;local&quot; in abstract Hilbert space with regard to some real-world set of quantum states, then it seems as though they would work equally well for a lot of the real world data that lives on a low-dimensional manifold in a high-dimensional space — the kind of data that machine learning (and eventually artificial general intelligence) seeks to tackle.

Talking about the benefits of dnn, what about the levels of abstraction? Each layer add levels of abstraction that you can&#x27;t see in shallow networks.

&gt; &quot;deep learning will not produce the universal algorithm&quot;<p>I doubt that a general algorithm exists (why should it?).<p>But well, if we are talking about human-level (or superhuman-level) AI, it is good to remember that WE are deep, recurrent neural networks (with a very different implementation, and spikes instead of floats, but still). If it work in vivo, why its abstracted version shouldn&#x27;t work in silico?

Entirely content-free post. Click-bait most likely.

&gt; Nothing is more frustrating when discussing deep learning that someone explaining their views on why deep neural networks are “modeled after how the human brain works” (much less true than the name suggests) and thus are “the key to unlocking true artificial intelligence”.<p>While I get what he is saying here, and more or less agree, I think it is not to be taken lightly that there <i>is</i> a significant difference in this discussion now as compared to 30 years ago. The difference is not <i>how</i> neural networks work, which clearly differs but is related in some ways to the brain, but rather <i>what</i> neural networks see.<p>What is really significant when you can handle lots and lots of data, and throw it all at a giant neural network, is what we see happening in the network. The observation that the hidden-layer filters developed as an optimal feature for classifying images appear to be Gabor-like directional filters (I&#x27;m referring of course to this type of thing [1]) is not random, and not an insignificant result. It really does relate to perception, in the sense that 1) we know that the brain has directional filters in the visual cortex and 2) more importantly, from signal processing theory we know that such filters are &quot;optimal&quot; from a certain mathematical point of view, and if they develop naturally as the best way to interpret &quot;natural&quot; images (or other natural data, such as audio [2]), it shows that development of such filters in the brain is perhaps also quite likely. There is quite some research in neuroscience at the moment looking for evidence of such optimal filters in early neural pathways.<p>So yes, neural networks are not models of &quot;how the brain works&quot;, but the newly established ability to process huge amounts of data, and to examine what kind of learning happens in order to optimise this processing, can tell us a lot about the brain -- not how it works, but what it must <i>do</i>. Complemented with work in neuroscience, the idea of modeling information processing is <i>not</i> unrelated and can really lead to some significant contributions in our understand of perception.. and perhaps, eventually, cognition -- but who knows.<p>The misunderstanding here is thinking that the be-all and end-all of neuroscience is studying how neurons fire and interact. Neuroscience is much more than that. Neuroscientists want to know how we experience and understand the world, and a big part of that is understanding what is required to process and interpret information, what is the information, what are its statistics, and what kind of neural processing would be required to extract it from our sensory inputs. Of course, this must be complemented by studies of how humans <i>do</i> react to stimuli, to try to verify that we <i>do</i> process information according to some model. But that model being verified -- that comes from what we know about information processing, and computer science can contribute there in a significant way.<p>[1]: <a href="https:&#x2F;&#x2F;computervisionblog.files.wordpress.com&#x2F;2013&#x2F;05&#x2F;gabor.png" rel="nofollow">https:&#x2F;&#x2F;computervisionblog.files.wordpress.com&#x2F;2013&#x2F;05&#x2F;gabor...</a><p>[2]: <a href="http:&#x2F;&#x2F;www.nature.com&#x2F;neuro&#x2F;journal&#x2F;v5&#x2F;n4&#x2F;abs&#x2F;nn831.html" rel="nofollow">http:&#x2F;&#x2F;www.nature.com&#x2F;neuro&#x2F;journal&#x2F;v5&#x2F;n4&#x2F;abs&#x2F;nn831.html</a>

Define your terms. WTF is &quot;Deap Learning&quot;?