CogAlg

Full introduction: <www.cognitivealgorithm.info>

Intelligence is a general cognitive ability, ultimately the ability to predict. That includes planning, which technically is a self-prediction. Any prediction is interactive projection of known patterns, hence the first step must be pattern discovery (more common but obfuscating term for that is unsupervised learning). These definitions are not terribly radical, pattern recognition is a core of any IQ test. But there is no conceptually consistent implementation, thus my design is from-the-scratch, although with many widely used elements.

For excellent popular introductions to cognition-as-prediction perspective see “On Intelligence” by Jeff Hawkins and “How to Create a Mind“ by Ray Kurzweil. On a technical level though, they and most everyone else use neural nets, which work in very coarse statistical fashion. Capsule Networks, recently introduced by Geoffrey Hinton et al, are more local and selective by multiple instantiation parameters. But they still start with weighted summation per parameter, which degrades the data before any comparison and evaluation.

In the next section, I define atomic comparison and resulting patterns, then describe a hierarchically recursive algorithm of search for incrementally complex patterns. The following sections compare my scheme to ANN, BNN, and CapsNet. This is an open project, we need help with design and implementation: WIKI. I pay for contributions or monthly if there is a track record, see CONTRIBUTING.

Outline of my approach

Proposed algorithm is a comparison-first alternative to deep learning, neither statistical nor neuromorphic. It's a search for patterns of incremental composition in recursively extended pipeline, a version of instance-based learning. Each level of my pipeline cross-compares inputs, then clusters them into patterns by resulting match. These patterns are inputs for the next level of search (cross-comparison), which forms patterns of patterns, etc. Higher levels also feed back average deviations of past match, combined across levels, to filter future inputs.

Pattern is a cluster of matching inputs, where match is a composite of compression per variable (vs. conventional per sequence). More commonly, pattern means a set of co-varying inputs, where proximate inputs don't match but their derivatives do. I call that a higher-derivation pattern, secondary to pattern and derivatives formed by cross-comparison of primary inputs. This is covered by my definition: input may consist of any set of variables. But cross-comparison should proceed with incremental distance between and derivation order of comparands.

First-level comparands are sensory inputs at the limit of resolution: adjacent pixels of video or equivalents in other modalities. All primary modalities output a dense array of such atomic inputs in Cartesian dimensions, and symbolic data is merely a product of their encoding by some cognitive process. To discover meaningful patterns, these symbols must be decoded before cross-comparison. The difficulty of decoding is exponential with the level of encoding, thus a start with raw sensory input is by far the easiest to implement.

This specific process description may seem like quite a jump from generalities in the first section. But it really isn’t, internally consistent pattern discovery requires that the process is strictly bottom-up, in complexity of both inputs and operations. And there is no ambiguity at the bottom: predictive value that defines patterns is match from comparison among their elements, initially those pixels. So, I think this process is uniquely consistent with my definitions, please let me know if you see any discrepancy in either.

####Comparison, more in part 1:

Basic comparison is inverse arithmetic operation between single-variable comparands, of incremental power: Boolean, subtraction, division, etc. Each order of comparison forms miss (loss): XOR, difference, ratio, etc., and match (similarity), which can be defined directly or as inverse deviation of miss. Direct match is compression of represented magnitude by replacing larger input with corresponding miss between the inputs: Boolean AND, the smaller input in comp by subtraction, integer part of ratio in comp by division, etc.

These direct similarity measures work if input intensity represents some stable physical property, which anti-correlates with variation. This is the case in tactile but not in visual input: brightness doesn’t correlate with inertia or invariance, dark objects are just as stable as bright ones. Thus, initial match in vision should be defined indirectly, as inverse deviation of variation in intensity. 1D variation is difference, ratio, etc., while multi-D comparison has to combine them into Euclidean distance and gradient, as in common edge detectors.

####Patterns, more in part 2:

Cross-comparison among patterns forms match and miss per parameter, as well as dimensions and distances: external match and miss (these are separate parameters: total value = precision of what * precision of where). Comparison is limited by max. distance between patterns. Overall hierarchy has incremental dimensionality: search levels ( param levels ( pattern levels)).., and pattern comparison is selectively incremental per such level. This is hard to explain in NL, please see the code, starting with line_patterns and line_PPs.

Resulting matches and misses are summed into lateral match and miss per pattern. Proximate input patterns with above-average match to their nearest neighbors are clustered into higher-level patterns. This adds two pattern levels: of composition and derivation, per level of search. Conditional cross-comp over incremental range and derivation, among the same inputs, may also add sub-levels in selected newly formed patterns. On a pixel level, incremental range is using larger kernels, and incremental derivation starts with using Laplacian.

####Feedback, more in part 3 (needs editing):

Average match is the first type of “input” filters, computed on higher levels. There are also “positional” filters, starting with pixel size and maximal pixel-comparison distance. Quantization order (bit, integer, float...) of such internal and external filters corresponds to the order of comparison, and their value is updated by feedback. These filters work similarly to “attention” in Neural Nets, but I have no equivalent of weight matrix. That’s because learning here is lateral clustering, vs. vertical clustering via backprop or Hebbian learning in ANN.

All filter types represent higher-level averages or co-averages, locally projected by higher-level patterns. Clustering on a filtered level is by the sign of cross-input-match deviation from those filters, so using averages balances positive and negative patterns (spans of above- and below- average cross-match) in future inputs. Such balanced filtering defines inputs that are likely to form novel patterns, avoiding both repetition of known higher-level patterns: predictable inputs, and forming long new negative-value patterns: noisy input spans.

####Hierarchy, part 4 is out of date:

Note that there is a single top-order hierarchy, with feedforward inputs and feedback filters passing through the same levels of search and composition. Lower orders are unfolded sequentially, so there’s always a single currently-unfolded hierarchy. That’s why I don’t have many diagrams: they are good at showing relations in 2D, but I have a simple 1D sequence of levels. Nested sub-hierarchies are generated by the process itself, depending on elevation in a higher-order hierarchy. That means I can’t show them in a generic diagram.

Brain-inspired schemes have separate sensory and motor hierarchies, in mine they combined into one. The equivalent of motor patterns in my scheme are positional filter patterns, which ultimately move the sensor. The first level is co-located sensors: targets of input filters, and more coarse actuators: targets of positional filters. I can think of two reasons they are separated in the brain: neurons and axons are uni-directional, and training process has to take the whole hierarchy offline. Neither constraint applies to my scheme.

Final algorithm will consist of first-level operations + recursive increment in operations per level. The latter is a meta-algorithm that extends working level-algorithm, to handle derivatives added to current inputs. So, the levels are: 1st level: G(x), 2nd level: F(G)(x), 3rd level: F(F(G))(x).., where F() is the recursive code increment. Resulting hierarchy is a pipeline: patterns are outputted to the next level, forming a new level if there is none. As long as there are novel inputs, higher levels will discover longer-range spatio-temporal and then conceptual patterns.

Please see system diagram.

• Some notes:

• There should be a unique set of operations added per level, hence a singular in “cognitive algorithm”.
• Core design should be done theoretically: generality requires large upfront investment in process complexity, which makes it look like overkill for any specific task. This is probably why such schemes are not explored.
• Many readers note disconnect between abstractions in this outline, and the amount of detail in current code. That’s because we are in space-time continuum: search must follow proximity in each dimension, which requires specific processing. It’s not specific to vision, the process is mostly the same for all raw modalities.
• Another complaint is that I don't use mathematical notation, but it doesn't have the flexibility to express deeply conditional process, with recursively increasing complexity.
• Most people who aspire to work on AGI think in terms behaviour and robotics. I think this is far too coarse to make progress, the most significant mechanisms are on the level of perception. Feedforward (perception) must drive feedback (action), not the other way around.
• Other unhealthy distractions are supervision and reinforcement. These are optional task-specific add-ons, core cognitive process is unsupervised pattern discovery, and main problem here is scaling in complexity.
• Don’t even start me on chatbots.

Comparison to Artificial and Biological Neural Networks

All unsupervised learning is some form of pattern discovery: input comparison and clustering. I do both laterally: among inputs within a level, while in statistical learning they are vertical: between layers of weighted summation. Weight adjustment by error in final comparison is effectively soft clustering: modulated inclusion or exclusion of subsequent inputs into next output. So, vertical weighted summation is primary to comparison, which I think is a conceptual flaw: local lateral cross-correlation should precede more distant vertical one.

Neural Nets is a version of statistical learning, from my perspective specifically a centroid clustering (centroid doesn’t have to be a single value, fitted line in linear regression can be considered a one-dimensional centroid). Basic ANN is a multi-layer perceptron: each node weighs the inputs at synapses, then sums and thresholds them into output. This normalized sum of inputs is their centroid. Output of the top layer is compared to some template, forming an error. Stochastic Gradient Descent then backpropagates the error, training initially random weights into transformations (reversed vertical derivatives) that reduce future error.

That usually means training CNN to perform some sort of edge-detection or cross-correlation (same as my cross-comparison but the former terms lose meaning on higher levels of search). But CNN operations are initially random, while my process is designed for cross-comp from the start. This is why it can be refined by my feedback, updating the filters, which is far more subtle and selective than weight training by backprop. So, I have several problems with basic process in ANN:

• Vertical learning (via feedback of error) takes tens of thousands of cycles to form accurate representations. That's because summation per layer degrades positional input resolution. With each added layer, the output that ultimately drives learning contains exponentially smaller fraction of original information. My cross-comp and clustering is far more complex per level, but the output contains all information of the input. Lossy selection is only done on the next level, after evaluation per pattern (vs. before evaluation in statistical methods).

• Both initial weights and sampling that feeds SGD are randomized. Also driven by random variation are RBMs, GANs, VAEs, etc. But randomization is antithetical to intelligence, it's only useful in statistical methods because they merge inputs with weights irreversibly. Thus, any non-random initialization and variation will introduce bias. All input modification in my scheme is via hyper-parameters, stored separately and then used to normalize inputs for comparison to inputs formed with different-value hyper-parameters.

• SGD minimizes error (top-layer miss), which is quantitatively different from maximizing match: compression. And that error is w.r.t. some specific template, while my match is summed over all past input / experience. The “error” here is plural: lateral misses (differences, ratios, etc.), computed by cross-comparison within a level. All inputs represent environment and have positive value. But then they are packed (compressed) into patterns, which have different range and precision, thus different relative value per relatively fixed record cost.

• Representation in ANN is fully distributed, similar to the brain. But the brain has no alternative: there is no substrate for local memory or program in neurons. Computers have RAM, so parallelization is a simple speed vs. efficiency trade-off, useful only for complex semantically isolated nodes. Such nodes are patterns, encapsulating a set of co-derived “what” and “where” parameters. This is similar to neural ensemble, but parameters that are compared together should be localized in memory, not distributed across a network.

More basic neural learning mechanism is Hebbian, though it is rarely used in ML. Conventional spiking version is that weight is increased if the synapse often receives a spike just before the node fires, else the weight is decreased. But input and output don't have to be binary, the same logic can be applied to scalar values: the weight is increased / decreased in proportion to some measure of similarity between its input and following output of the node. That output is normalized sum of all inputs, or their centroid.

Such learning is local, within each node. But it is still a product of vertical comparison: centroid is of a higher order of composition than individual inputs. This comparison across composition drives all statistical learning, but it destroys positional information at each layer. Compared to autoencoders: principal backprop-driven unsupervised learning technique, Hebbian learning lacks the decoding stage (as does the proposed algorithm). Decoding decomposes hidden layers, to equalize composition orders of output and compared template.

Inspiration by the brain kept ANN research going for decades before they became useful. Their “neurons” are mere stick figures, but that’s not a problem, most of neuron’s complexity is due to constraints of biology. The problem is that core mechanism in ANN, weighted summation, may also be a no-longer needed compensation for such constraints: neural memory requires dedicated connections. That makes representation and cross-comparison of individual inputs very expensive, so they are summed. But we now have dirt-cheap RAM.

Other biological constraints are very slow neurons, and the imperative of fast reaction for survival in the wild. Both favor fast though crude summation, at the cost of glacial training. Reaction speed became less important: modern society is quite secure, while continuous learning is far more important because of accelerating progress. Summation also reduces noise, which is very important for neurons that often fire at random, to initiate and maintain latent connections. But that’s irrelevant for electronic circuits.

I see no way evolution could produce proposed algorithm, it is extremely limited in complexity that can be added before it is pruned by natural selection. And that selection is for reproduction, intelligence here is distantly instrumental. The brain evolved to guide the body, with neurons originating as instinctive stimulus-to-response converters. Hence, both SGD and Hebbian learning is fitting, driven by feedback of action-triggering weighted input sum. Pattern discovery is their instrumental upshot, not an original purpose.

Uri Hasson, Samuel Nastase, Ariel Goldstein reach a similar conclusion in “Direct fit to nature: an evolutionary perspective on biological and artificial neural networks”: “We argue that neural computation is grounded in brute-force direct fitting, which relies on over-parameterized optimization algorithms to increase predictive power (generalization) without explicitly modeling the underlying generative structure of the world. Although ANNs are indeed highly simplified models of BNNs, they belong to the same family of over-parameterized, direct-fit models, producing solutions that are mistakenly interpreted in terms of elegant design principles but in fact reflect the interdigitation of ‘‘mindless’’ optimization processes and the structure of the world.”

Comparison to Capsule Networks

The nearest experimentally successful method is recently introduced “capsules”. Some similarities to CogAlg:

• capsules also output multivariate vectors, “encapsulating” several parameters, similar to my patterns,
• these parameters also include pose: coordinates and dimensions, compared to compute transformations,
• these transformations are compared to find affine transformations or equivariance: my match of misses,
• capsules also send direct feedback to lower layer: dynamic routing, vs. trans-hidden-layer backprop in ANN.

My main problems with CapsNet and alternative treatment:

• Object is defined as a recurring configuration of different parts. But such recurrence can’t be assumed, it should be derived by cross-comparing relative position among parts of matching objects. This can only be done after their positions are cross-compared, which is after their objects are cross-compared: two levels above the level that forms initial objects. So, objects formed by positional equivariance would be secondary, though they may displace initial segmentation objects as a primary representation. Stacked Capsule Autoencoders also have exclusive segmentation on the first layer, but proximity doesn’t matter on their higher layers.

• Routing by agreement is basically recursive centroid clustering, by match of input vector to the output vector. The output (centroid) represents inputs at all locations, so its comparison to inputs is effectively mixed-distance. Thus, clustering in CapsNet is fuzzy and discontinuous, forming redundant representations. Routing by agreement reduces that redundancy, but not consistently so, it doesn’t specifically account for it. My default clustering is exclusive segmentation: each element (child) belongs to only one cluster (parent). Fuzzy clustering is selective to inputs valued above the cost of adjusting for overlap in representation, which increases with the range of cross-comparison. This conditional range increase is done on all levels of composition.

• Instantiation parameters are application-specific, CapsNet has no general mechanism to derive them. My general mechanism is cross-comparison of input capsule parameters, which forms higher-order parameters. First level forms pixel-level gradient, similar to edge detection in CNN. But then it forms proximity-constrained clusters, defined by gradient and parameterized by summed pixel intensity, dy, dx, gradient, angle. This cross-comparison followed by clustering is done on all levels, with incremental number of parameters per input.

• Number of layers is fixed, while I think it should be incremental with experience. My hierarchy is a dynamic pipeline: patterns are displaced from a level by criterion sign change and sent to existing or new higher level. So, both hierarchy of patterns per system and sub-hierarchy of derivatives per pattern expand with experience. The derivatives are summed within a pattern, then evaluated for extending intra-pattern search and feedback.

• Output vector of higher capsules combines parameters of all lower layers into Euclidean distance. That is my default too, but they should also be kept separate, for potential cross-comp among layer-wide representations. Overall, CapsNet is a variation of ANN, with input summation first and dynamic routing second. So, it’s a type of Hebbian learning, with most of the problems that I listed in the previous section.

Atomic comparison: quantifying match and miss between variables

First, we need to quantify predictive value. Algorithmic information theory defines it as compressibility of representation, which is perfectly fine. But compression is currently computed only for sequences of inputs, while I think a logical start is analog input digitization: a rock bottom of organic compression hierarchy. The next level is cross-comparison among resulting pixels, commonly known as edge detection, and higher levels will cross-compare resulting patterns. Partial match computed by comparison is a measure of compression.

Partial match between two variables is a complementary of miss, in corresponding power of comparison:

• Boolean match is AND and miss is XOR (two zero inputs form zero match and zero miss),
• comparison by subtraction increases match to a smaller comparand and reduces miss to a difference,
• comparison by division increases match to min * integer part of ratio and reduces miss to a fractional part (direct match works for tactile input. but reflected-light in vision requires inverse definition of initial match)

In other words, match is a compression of larger comparand’s magnitude by replacing it with miss. Which means that match = smaller input: a common subset of both inputs, = sum of AND between their uncompressed (unary code) representations. Ultimate criterion is recorded magnitude, rather than bits of memory it occupies, because the former represents physical impact that we want to predict. The volume of memory used to record that magnitude depends on prior compression, which is not an objective parameter.

Given incremental complexity, initial inputs should have binary resolution and implicit shared coordinate (being a macro-parameter, resolution of coordinate lags that of an input). Compression of bit inputs by AND is well known as digitization: substitution of two lower 1 bits with one higher 1 bit. Resolution of coordinate (input summation span) is adjusted by feedback to form integers that are large enough to produce above-average match.

Next-order compression can be achieved by comparison between consecutive integers, distinguished by binary (before | after) coordinate. Basic comparison is inverse arithmetic operation of incremental power: AND, subtraction, division, logarithm, and so on. Additive match is achieved by comparison of a higher power than that which produced comparands: comparison by AND will not further compress integers previously digitized by AND.

Rather, initial comparison between integers is by subtraction, resulting difference is miss, and absolute match is a smaller input. For example, if inputs are 4 and 7, then miss is 3, and their match or common subset is 4. Difference is smaller than XOR (non-zero complementary of AND) because XOR may include opposite-sign (opposite-direction) bit pairs 0, 1 and 1, 0, which are cancelled-out by subtraction.

Comparison by division forms ratio, which is a magnitude-compressed difference. This compression is explicit in long division: match is accumulated over iterative subtraction of smaller comparand from remaining difference. In other words, this is also a comparison by subtraction, but between different orders of derivation. Resulting match is smaller comparand * integer part of ratio, and miss is final reminder or fractional part of ratio.

A ratio can be further compressed by converting to a radix | logarithm, and so on. But computational costs may grow even faster. Thus, power of comparison should increase only for inputs sufficiently compressed by lower power: AND for bit inputs, SUB for integer inputs, DIV for pattern inputs, etc. Actual compression depends on input and on resolution of its coordinate: input | derivative summation span. We can’t control the input, so average match is adjusted via coordinate resolution.

But the costs of operations and incidental sign, fraction, irrational fraction, etc. may grow even faster. To justify the costs, the power of comparison should only increase in patterns of above-average match from prior order of comparison: AND for bit inputs, SUB for integer inputs, DIV for pattern inputs, etc. A filter for inclusion into positive or negative pattern is a past match that co-occurs with average higher-level projected match. Relative match is accumulated until it exceeds the cost of updating lower-level filter, which terminates same-filter span.

To filter future inputs, this absolute match should be projected: recombined with co-derived miss projected for a target distance. Filter deviation is accumulated until it exceeds the cost of updating lower-level filter. Which then forms relative match: current match - past match that co-occurs with average higher-level projected match. This relative match: above- or below- average predictive value, determines input inclusion into positive or negative predictive value pattern.

Separate filters are formed for each type of compared variable. Initial input, such as reflected light, is likely to be incidental and very indirectly representative of physical properties in observed objects. Then its filter will increase, reducing number of positive patterns, potentially down to 0. But differences or ratios between inputs represent variation, which is anti-correlated with match. They have negative predictive value, inverted to get incrementally closer to intrinsically predictive properties, such as mass or momentum.

Absent significant correlation between input magnitude and represented physical object magnitude, the only proxy to match in initial comparison is inverse deviation of absolute difference: average_|difference| - |difference|. Though less accurate (defined via average diff vs. individual input), this match is also a complementary of diff:

• complementary of |difference| within average_|difference| (=max of the |difference| s), similar to minimum:
• complementary of |difference| within max input.

quantifying lossy compression

There is a general agreement that compression is a measure of similarity, but no one seems to apply it from the bottom up, the bottom being single scalars. Also, any significant compression must be lossy. This is currently evaluated by perceived similarity of reconstructed input to the original input, as well as compression rate. Which is very coarse and subjective. Compression in my level of search is lossless, represented by match on all levels of pattern. All derived representations are redundant, so it’s really an expansion vs. compression overall.

The lossy part comes after evaluation of resulting patterns on the next level of search. Top level of patterns is cross-compared by default, evaluation is per lower level: of incremental derivation and detail in each pattern. Loss is when low-relative-match buffered inputs or alternative derivatives are not cross-compared. Such loss is quantified as the quantity of representations in these lower levels, not some subjective quality.

Compression also depends on resolution of coordinate (input summation span), and of input magnitude. Projected match can be kept above system’s average by adjusting corresponding resolution filters: most significant bits and least significant bits of both coordinate and magnitude.

Implementation

Any prediction has two components: what and where. We must have both: value of prediction = precision of what * precision of where. That “where” is currently neglected: statistical ML methods represent coordinates much more coarsely than the inputs. Hence, precision of where (spans of and distances between patterns) is degraded, and so is predictive value of combined representations. That's not the case here because my top-level patterns (multi-dimensional blobs) are contiguous.

Core algorithm is 1D: time only. Our space-time is 4D, and average match is presumably equal over all dimensions. That means patterns defined in fewer dimensions will be only slices of actual input, fundamentally limited and biased by the angle of scanning / slicing. Hence, initial pixel comparison should also be over 4D at once, or at least over 3D for video and 2D for still images. This full-D-cycle level of search is a universe-specific extension of core algorithm. The dimensions should be discoverable by the core algorithm, but coding it in is much faster.

This repository currently has three versions of 1st D-cycle, analogous to connected-component analysis: 1D line alg, 2D frame alg, and 3D video alg. Subsequent cycles will compare full-D-terminated input patterns over increasing distance in each dimension, forming discontinuous patterns of incremental composition and range. “Dimension” here defines external sequence and distance among inputs. This is different from conventional clustering, which treats both external and internal parameters as dimensions.

Complete hierarchical algorithm will have two-level code:

• 1st level algorithm: contiguous cross-comparison over full-D cycle, plus feedback to adjust most and least significant bits of the input.
• Recurrent increment in complexity, extending current-level alg to next-level alg. This increment will account for increasing internal complexity of input patterns on higher levels, unfolding them for cross-comparison and re-folding results for evaluation and feedback.

Initial testing could be on recognition of labeled images, but video or stereo video should be much better. We will then add colors, maybe audio and text.

For more detailed account of current development see WIKI.

Suggestions and collaboration are most welcome, see CONTRIBUTING.