Denoising Dense Y (part 1): Permeability of Reconfiguration

What noise did in an information system? Is it possible to modulate noise? What is denoising?

The dense noise resembles the multiple independent images in one frame with various complexities. The frame which is also the metaphor for the medium of transmission is unable to accommodate their collision, affecting network space that is initially as an infinite entity to be saturated in transmission exchange while the mediators dealing with the chaotic movement from overload input. Then, the information must be connected to generator to filter out the noise, so the receiver can receive a “clean information”.

(image 1: banality or digital accident in google maps, captured on Nov 19, 2020)

However, the noise through filtering process has been eliminated or even called as intruders since its capability for impoverishing phenotype information (influenced by the technical environment), but also signalling new information outside the phenotype variable with preconditioning noise occurs in a set complex of collective environment. We can look, as the general example, the collision between transmission on the sharing medium while each transmission is bearing various sensory variables, hence the amount of information we receive/sense depends on the occurred probabilities including its noise. Thus, are there any ways to denoise? Denoising or in the context of image denoising can be mathematically written as:

y=x+ny

y is an observed noisy image; x is an unknown clean image, and n is a representation of additive white Gaussian noise[1], but n is replaced into variables representation and y as dense noise to broaden our understanding in the epistemological framework using digital logic; the equation becomes:

y= x+n

Yet, mathematics or even this equation does not adequately represent or become a generator of noise and order[2] since noise tends to breach sporadically from the established system and find its way to an alternative ecosystem. If obliterating or trying to cover noise is unfeasible, denoising offering us to comprehend noise as information without restoring it to phenotype information. Denoise is not an escape system or transcendental medium for noise before it propagates on the medium. De- attempts to reload noise information in a form and interpret it both in metaphorical, symbolic, and technical terms without a set of semantic configurations. After all, noise can be perceived not as information resistance but as new information with its discrete form. If we held this premise, what if denoising creates a new noisy image that is often associated with a digital accident or a banal one instead of adhering to initial form?

Permeability of Reconfiguration

The way technology operates mostly considers noise in information transmission, often interfering with and showing us other signals, which can be arbitrary interpreted as distraction and diversion of phenotype information. The term “phenotype information” will be used here to describe material information that sensory perceivable and being processed through pre-requisite imaginary and conditioning of environmental variable; a contrast to the preliminary information that is more intuitive by nature and constitutes of pre-conditioning. Preliminary information is the information to be conveyed and as pure material of communication system. In zoom meeting, they type of image disturbance we often experience is an intruder-unwanted object entering the frame can be our cat, people mill about, or unidentified shadow, but with a note that the sender does not intend to include these objects in processed information and the receiver does not expect to see or experience it. The sender intended to send the preliminary information, and in exchange what the receiver receives is the phenotype information.

There are hope and expectation in information transmission that make a whole process a possible open system for conditioning phenotype information; thus, information system and communication always put noise into account and maximise signal ratio to noise to obtain ideal message transmission which close to preliminary information although noise is impossible to annihilate[3]. Meanwhile, the receiver, as the open possibilities holder, has been also alienated from the coding system; but this status quo is actually quite beneficial as it allows them to have the control interpreting the received information and recognising the information through its sequence. It is on receiver whether the additional or foreign objects in image sequence are categorised as noise or not.

Noise affects information in a more extensive open system [4], carrying patterns and randomness in one repeated medium; so that noise can be a material of information and independent from semantic interpretations. It is neither positive nor negative nor null since noise does not precede or to be preceded. Noise appears as an error forcing and demanding a change given by the instability in an already coded system with specific calculation and probabilities that are often seen as entropy; nonetheless, noise due to its characteristic to the system is a paradoxical entity: system in system, a reservoir for new information and elements that are continuously infiltrating, betraying its system, sporadic yet require system as a propagator for its discrete form. This dynamic character mediates system and randomness and becomes an oscillator for the probability of information transmission, causing noise to open the virtualities and uncertainties. Thus, noise no longer shows as corrupted data, digital accidents, or errors but it can generate knowledge production and a new type of information.

(diagram reconfiguration of permeability, courtesy: Rafika Lifi)

Borrowing the autoencoder process diagram [5], dense noise (y) is represented in diagram. y has a set y1-y4 with a layered system, y’ is denoising result with a set array of 4,3,2,1, and x is repressed representation from the reduction of data Y producing model Y’; so mathematically we get z= y + y’. We can expand this ideal representation to y is the noise field, and each layer 1–4 consist of a diverse range in the amount, type, and character of information. Each digit is also bearing the system’s set from a-z and every ‘a’ has component probabilities a1 and its multiple set. The condition will be intertwined into a network trace although is unstructured and does not occupy a fixed system causing a porous network which its gap ready to be infiltrated; thus, y will not frequently produce y’, and x becomes temporary propagation system. Noise is a trace laden within its different degree of complexity, making the permeability possible after the system that is often conditioned for homogeneous information reception get challenged by different degrees of the propagation medium.

To synthesise x, the dense y must be reduced including each multilayered data dimension before it creates an entire unit system to get the given random system as a synthesis or functional unit of the variable of process. In the meantime, this function determines the probability structure of a given random system depends on the systematic scheme of technique, time, and space while the causality will always occur between the process; thus y ≠ y’. The system also offers dynamic progress based on the advancement of the phenotype variable; even isolated noise bears a unique sequence of conditioning.

Through maintaining its environment away from systemic equilibrium, noise can identify and characterise itself as dynamic progress and metastasis. Noise, as well as denoising, indicates the emerge of discrete form and also the dynamic progress of a domain, showing us the transformation of the field whose its current condition can create a new autonomous entity that adapts with the original environment since noise will still be affected by the phenotype characteristic in receiver stage. This denoising shows us the process of new formation and transformation in one domain and conceptual changes from one process to another. Moreover, despite our consciousness on noisy image, it is our alienation from digital logic that leads us to iteratively capture, label, and comprehend the information system. Denoising in turn not demisting noise to be fit in phenotype information; it acknowledges noise as a reservoir of novel information rejecting the established system, growing as a threat while offering freedom of choice and new possibilities.

[1] Additive White Gaussian Noise (AWGN) is random noise value with a given distribution to model of noise
[2] Refers to Lem’s statement in Summa Technologiae “mathematics is not a noise generator. It is a generator of order,of various “internal orders.”” However, in the noise case, order is used as a temporary propagator of the previous environment. Mathematics is not sufficient to represent either noise or order of noise due to the nature of noise.
Lem, S. (2014). Summa Technologiae, trans. Joanna Zylinska. p.375
[3] Shannon’s communication theory provides an establishing technical framework on noise by designing systems that suppress as much noise as possible while recognising that information transmission cannot be transmitted without noise. Noise is considered to be a chance variable that arises through a stochastic (random) process in an open and more complex system.
Shannon, C. E. (1948). A mathematical theory of communication. The Bell system technical journal, 27(3), 379–423.
[4] Idem., 19
[5] Autoencoder is an unsupervised neural network model to learn a compressed representation of an input and how to reconstruct the data.

Part 2 will elaborate individuation of noise and the possibilities of aesthetics transpositioning in neural network