Composing Popular Music with Physarum polycephalum-based

Composing Popular Music with Physarum

polycephalum-based Memristors

Satvik Venkatesh

Interdisciplinary Centre for

Computer Music Research

University of Plymouth

United Kingdom

satvik.venkatesh

@plymouth.ac.uk

Edward Braund

Interdisciplinary Centre for

Computer Music Research

University of Plymouth

United Kingdom

edward.braund

@plymouth.ac.uk

Eduardo Reck Miranda

Interdisciplinary Centre for

Computer Music Research

University of Plymouth

United Kingdom

eduardo.miranda

@plymouth.ac.uk

ABSTRACT

Creative systems such as algorithmic comp osers often use

Artiﬁcial Intelligence models like Markov chains, Neural

Networks, and Genetic Algorithms to model stochastic pro-

cesses. Unconventional Computing (UC) technologies ex-

plore non-digital ways of data storage, processing, input,

and output. UC paradigms such as Biocomputing and Quan-

tum Computing delve into domains beyond the binary bit

to handle complex non-linear functions. In this pap er, we

harness Physarum polycephalum as a memristor to process

and generate creative data for p opular music. The organism

works as a collaborator in the process of composing our song

titled Creep into my Lawn. While there has been research

conducted in this area, the literature lacks examples of pop-

ular music and how the organism’s non-linear behaviour can

b e controlled while composing music. This is important be-

cause non-linear forms of representation are not as obvious

as conventional digital means. This study aims at dissemi-

nating this technology to non-experts and musicians so that

they can incorporate it in their creative processes. Further-

more, it combines resistors and memristors to have more

ﬂexibility while generating music and optimises parameters

for faster processing and performance.

Author Keywords

Unconventional Computing, Biocomputing, Musical Com-

p osition, Popular Music, Physarum polycephalum

CCS Concepts

•Theory of computation → Unconventional Com-

puting; •Applied computing → Sound and Music

Computing; Algorithmic Composition;

1. INTRODUCTION

There is a vast corpus of research that explores computa-

tional processes to exhibit creative behaviour [8]. The pro-

cess of musical composition frequently involves two elements

— material that has been previously heard and ideas that

are new and innovative [9]. The former encompasses char-

acteristics of the composer’s style and their inspirations.

The latter has been explored in computational creativity

Licensed under a Creative Commons Attribution

4.0 International License (CC BY 4.0). Copyright

remains with the author(s).

NIME’20, July 21-25, 2020, Royal Birmingham Conservatoire,

Birmingham City University, Birmingham, United Kingdom.

through stochastic processes, which aim to induce random-

ness in the output of the system [19].

Many Artiﬁcial Intelligence (AI) models such as Markov

chains [12], Genetic Algorithms (GA), Artiﬁcial Neural Net-

works (ANNs) [23] have been successfully adopted to ren-

der creative systems. Markov chains compute a conditional

probability distribution of future states depending on the

current state [19]. The architecture of ANNs is inspired by

the structure of the brain and enables us to model com-

plex non-linear functions. Additionally, other mode ls such

as self-similarity and cellular automata [14], that do not

strictly fall under the AI bracket, have been implemented

to render creative systems [10].

Physical, chemical, and biological processes have not only

inspired AI algorithms, but given rise to other forms of com-

putation. Unconventional C omputing (UC) aims to develop

new computer architectures for data storage and processing

[1]. Computational creativity has been conventionally ren-

dered on classical computers based on the von Neumann

architecture. The fundamental unit of processing and stor-

age is the binary bit, that is either 0 or 1. Conversely, UC

technologies explore analog forms of computation and stor-

age. For example, quantum computing uses qubits as the

fundamental unit of processing.

Physarum polycephalum is a plasmodial slime mould that

has been widely harnessed for biocomputing applications.

Researchers have utilised its ability to optimise its shape ac-

cording to the environment to solve mazes [16], build logic

gates [2], and ﬁnd shortest paths [17]. Another interesting

prop erty of the organism is that it can act as a memristor

[11]. Memristor (a portmanteau of memory and resistor)

is the fourth fundamental passive circuit element, the ﬁrst

three being resistor, capacitor, and inductor. It was theo-

retically demonstrated by Leon Chua in 1971 [7], but was

physically realised in 2008 [22].

Research i nto memristor technology is gaining traction

due to many reasons. It combines storage and process-

ing into one unit and therefore, has the potential to opti-

mise computing architectures. [21, 13] demonstrated spike-

timing-dep endent plasticity (STDP)-like behaviour using

memristors. STDP has been adopted in feed-forward ANNs

for pattern recognition algorithms [3].

Physarum polycephalum-based memristors serve as an in-

expens ive and accessible UC prototype, when compared to

quantum computers and chemically manufactured memris-

tors. [5] designed 3D printed receptacles to enclose the or-

ganism into a portable electrical component. [6] presented a

compact kernel to interface with biomemristors for creative

practitioners and D.I.Y. prototyping enthusiasts. Moreover,

research is progressing with regards to making UC tech-

nologies available to individuals outside laboratories. This

514

makes it increasingly important to use these bespoke sys-

tems in creative works.

This pap er explores how Physarum polycephalum-based

memristors can be used in the process of composing popular

music. It works with Physarum polycephalum as a collabo-

rator and is a step towards the long-term dream o f building

a rock band of biomemristors. As we are using a non-linear

comp onent as the processing device, input parameters that

govern the composition process require non-digital forms

of representation. This study renders a practice-based ap-

proach towards understanding non-linear input param eters,

which m ay not be as obvious as binary logic. It aims to

break down the creative process and explores how these pa-

rameters inﬂuence the output of the PhyBox. Biocomputer

Rhythms by Eduardo Miranda was the ﬁrst musical piece

for piano and percussion composed with the PhyBox [15].

However, it was not composed in a popular music context

and there have been recent developments with mapping pro-

cedures, such as resistors can be used alongside memristors

in the process of composition [24]. This paper incorporates

these developments, optimises parameters to minimise pro-

cessing time, and oﬀers more degree of freedom to the user.

2. SYSTEM OVERVIEW

2.1 Hardware-wetware Framework

We maintain a Physarum polycephalum farm [1] grown in

plastic lunch boxes as shown in ﬁgure 1. In order to grow

Physarum polycephalum-based memristors and make them

a part of the composition process, we transfer a section of

the organism into receptacles [5]. These receptacles allow

the organism to act as an electrical component and are oper-

ated through the PhyBox [6], a portable kernel for creative

applications as shown in ﬁgure 2. PhyBox is powered by a

Raspbe rry Pi model B.

Figure 1: Physarum polycephalum culture grown in

a lunch box.

Training data of music is converted into voltages with the

help of a mapping procedure [24]. In this paper, training

data refers to the music originally composed by the author.

These voltages are serially applied across the memristors as

shown in ﬁgure 3. The current ﬂowing through the memris-

tor depends on its resistance (or memristance). The current

is again translated to music through a similar mapping pro-

cedure. Therefore, the memristors create new music based

on the training data.

Music is fed into the PhyBox through the MIDI proto-

col. MIDI data is broken down into four parameters —

pitch, velocity, time between note-ons (rhythm), and dura-

Figure 2: Setting up the PhyBox for musical com-

position.

tion. Each memristor processes one of these parameters in

parallel, that is there are 4 memristors in the circuit.

2.2 Mapping procedure

The framework for generating creative data harnesses the

spiking behaviour of memristors. Each time there i s a volt-

age change across the memristor, there is a spike in current

as shown in ﬁgure 3. This behaviour is diﬀerent from that of

linear components like resistors in electronic circuits. The

following explains its behaviour. When the positive change

in voltage is produced across the memristor, the resistance

is initially low. With time, it develops resistance and there

is a gradual decrease in the current ﬂowing through the cir-

cuit.

Figure 3: Spike generated by the memristor on ap-

plying four voltage changes — 0.75V, 1.5V, 0.5V,

and 1.5V. Positive spikes are produced for positive

voltage changes and vice versa.

2.2.1 Music to Voltages

In this system, these voltage transitions are attributed to

musical note transitions. A musical note is assigned to a

sp eciﬁc voltage on a priority basis. Musical notes that occur

more often are assigned higher priority and vice versa. If

there are two events with the same number of occurrences,

the more recent event is given a higher priority. This is

analogous to the learning rule in STDP, where the new event

is good and the old event is bad.

As we are using a Raspberry Pi, the voltage range is 0 to

3V. It is assigned by equation 1 [24].

515

(p −

)V olts (1)

where v

is the assigned voltage, N is the numb er of musical

notes and p is the priority of the event. A relationship

b etween popularity and assigned voltages is devised by the

following analogy. The rate of change of current (or slope)

in the beginning of the graph is greater when compared to

the end. Moreover, taller spikes are generated for greater

changes in voltage. The stability of resistance is related to

the popularity of the e vent. Therefore, more popular events

are mapped to lower voltages and less popular events are

mapp ed to higher voltages.

Note that when the training data is initially fed into the

PhyBox, there are no events in the memory. It is trained

during the course of the song. Therefore, the number of

variations that can possibly occur gradually increases with

more data.

2.2.2 Current to Music

A similar procedure is followed while converting current

readings to music. The voltage range is well-deﬁned to be 0

to 3V. However, a discrete current range is not deﬁned be-

cause the resistance of the memristor is unknown. Empiri-

cally, studies have approximated the resistance in the order

of mega-ohms [6]. Let I

max

and I

min

b e the maximum and

minimum current sensed during the spike respectively. The

current range [I

, I

] is deﬁned by the following equation.

, I

] =

min(V

)

min

)

max(V

)

max

)

(2)

where V

is the set of all assigned voltages and v

the input

voltage at that speciﬁc moment in time. The current range

is di vided into equal divisions and converted to music based

on midrise quantisation [4]. Again, the events are ordered

according to priority, where lower current values stand for

more popular events and vice versa.

2.2.3 Resistors

One of the advantages of this mapping procedure is that

it provides a way to incorporate resistors in the process of

musical composition. It deﬁnes a resistor’s behaviour as lin-

ear. Therefore, when the training data is fed into a resistor,

the output of the system is exactly identical to the input.

This oﬀers creative pathways to explore in the process of

comp osition, such as the rhythm can be identical to the in-

put and the pitch can be composed by the memristor. It is

imp ortant to note that the value of the resistance does not

inﬂuence the output. In this study, we used 330kΩ resistors.

2.3 Parameters

There are two parameters that monitor the output produced

by the system — dwell time and measurement oﬀset. Dwell

time speciﬁes how long the voltage is applied across the

memristor. This basically determines the processing time

of the memristors for each musical event. For example, if

the dwell time is 1 s, the memristor takes 20 s to process

20 musical events. The measurement oﬀset determines at

what point of time is the current value calculated in the

circuit. For instance, on applying 1.5 V at 0.1 s in ﬁgure

3, 0% oﬀset considers the current at 0.1 s and 100% oﬀset

considers the current at 1.1 s (Here the dwell time was 1

s). [6, 24] have conducted some experiments to understand

how these non-linear parameters inﬂuence the output of the

memristors. 0% oﬀset produces smaller deviations from the

input data when com pared to 100% oﬀset [24]. However,

there is still not a clear delineation of these functions due

to their complex non-linear behaviour.

3. EXPERIMENTS

In this paper, we us e Physarum polycephalum as a collabo-

rator in the process of musical comp osition. As the current

interface only accepts MIDI data, we have harnessed it to

comp ose parts for piano, drums, and the vocal melody. The

dwell time was constant at 1 s. The measurement oﬀset was

altered between 0%, 30%, 50%, and 100%.

3.1 Structure

The time and key signature of the song is 4/4 and E mi-

nor respectively. The genre of the song falls under Rock.

The song starts with the main riﬀ played only on the pi-

ano, which converges into the other instruments joining the

main riﬀ. Followed by that, we have one verse, pre-chorus,

and chorus, which are sung by a vocalist. The chorus has

the main riﬀ in the background. After the chorus, there

is a piano solo followed by a drum solo. The song ﬁn-

ishes with another chorus at the end. This song follows

the structure adopted by most popular music compositions.

However, it was slightly altered, for instance omitting the

second verse, in order to make things more interesting in

the view of Physarum polycephalum.

In order to explain the structure of the song with clar-

ity, the verse was divided into two parts. While feeding in

data for piano and drums, we divided the whole song into

3 blocks. This was done to avoid overloading the model

with too m any possibilities and transitions. The ﬁrst block

consisted of the intro and ﬁrst part of the verse. The sec-

ond block comprised the second part of the verse and the

pre-chorus. The third block contained the remainder of the

song. These blocks were not divided with regards to time,

but were divided with respect to number of MIDI events

and structure of the song. For instance, chords in the verse

contained more MIDI events than melody lines. The vocal

melo dy was transcribed as a MIDI ﬁle. MIDI events like

sustain pedal was removed from the training data.

3.2 Main Riff

The main riﬀ of the song is 4 bars in length. A piano tune

is backed up by drums and guitars for the main riﬀ. The

ﬁrst two bars of the piano melody were composed by the

author. We wanted the third and the fourth bar to follow a

similar pattern as the ﬁrst two, but with minor variations.

The aim is to create a feeling of repetition alongside some

anticipation in the listener. We conducted experiments with

measurement oﬀset values of 0%, 30%, and 100%. The gui-

tars and drums were composed by keeping the piano track

as a reference.

3.3 Verse

The instrumentation for the verse consisted of vocals, pi-

ano, and drums. The ﬁrst part of the verse had simple pi-

ano chords at the beginning of each bar. We experimented

with measurement oﬀset values of 0%, 30%, and 100%. The

objective of this experiment was to make Physarum poly-

cephalum improve the chord pattern, chord inversions, and

the voicing.

The second part of the verse had arpeggios played on the

piano. For this section of the song, we felt that there was less

ﬂexibility with altering the rhythm. We wanted to maintain

the rhythmic structure originally composed by the author.

Therefore, we attached m emristors for pitch and velocity

and resistors for time between not e-ons and duration. The

measurement oﬀset was set to 100% and this conﬁguration

was refe rred to as 100R.

3.4 Piano Solo

516

The harmonic and rhythmic complexity of the piano solo is

higher than the other sections of the song. The piano solo

was originally composed by the author. Also, there was a

scale shift within the solo from E minor to E Phrygian and

then C major. At the end of the piano solo, the song re-

turns to E minor. We wanted to explore how Physarum

po lycephalum would respond to scale changes and complex

melo dic ideas. This is the most interesting section of the

song because it provides a high degree of freedom with re-

gards to musical options. Additionally, it does not restrict

Physarum polycephalum with respect to the song’s instru-

mentation because the ot her instruments do not play during

the piano solo. Tests were conducted for measurement oﬀ-

sets of 5%, 30%, 50%, 100%, and 100%R.

3.5 Drums

We wanted to harness the organism to create new drum

ﬁlls based on existing ones. Additionally, we wanted to

explore its creativity for the drum solo. We tested it with

measurement oﬀset values of 0%, 30%, 100%, and 100%R.

3.6 Vocal Tune

A vocal melody was composed by the author and tran-

scribe d as a MIDI track. Our objective of using Physarum

po lycephalum for the vocal tune was to make it sound more

catchy and less monotonous. We explored measurement oﬀ-

sets of 0%, 30%, 100%, and 100%R.

3.7 Dissimilarity

In order to objectively calculate the diﬀerence between orig-

inal melody and Physarum polycephalum’s rendition, we de-

ﬁned a metric called dissimilarity, which is the number of

musical events that are diﬀerent from the original input.

Each of the parameters — pitch, velocity, time between

note-ons, and duration would return a value of dissimilarity.

The total dissimilarity is deﬁned as the s um of individual

dissimilarities of the four parameters.

4. RESULTS

All audio and MIDI ﬁles that were composed by the au-

thor and Physarum polycephalum can be found in this link

The naming format for the ﬁles is {block

Instrument Type

Oﬀset}. The title of the song is Creep into my Lawn. A mu-

sic video that illustrates the parts composed by Physarum

p olycephalum and the author is available in this link

. The

mo del used to generate musical data only uses note transi-

tions. This gives high ﬂexibility to Physarum polycephalum,

but opens up possibilities like time signature mismatch and

disharmonious chords. Therefore, few modiﬁcations were

made by the composer to suit the song. As the model is

based on musical transitions, in Physarum polycephalum’s

output, the ﬁrst note will be missing because no transition

has occurred yet. Hence, while evaluating its output, the

ﬁrst note was added from the input.

4.1 Dissimilarity

Table 1 contains the total dissimilarities for diﬀerent values

of measurement oﬀset. Consistently, 0% was more similar

to the input when compared to other oﬀsets. 30% was closer

to the input when compared to 1 00%, but the diﬀerence was

much smaller.

4.2 Main Riff

https://github.com/satvik-venkatesh/

Physarum-Music

https://youtu.be/dKQhnlO7ZrI

The main riﬀ originally composed by the author is shown in

ﬁgure 4. Physarum polycephalum’s output for measurement

oﬀset of 0% was not interesting. It resembled the origi-

nal melody except for changing one crotchet to a dotted

crotchet. For measurement oﬀsets of 30% and 100%, the

generated output sounded more appropriate for the main

riﬀ. The authors felt that 30% oﬀset had more resem-

blance to the original tune when compared to 100%. We

chose 100% and made no modiﬁcations to Physarum poly-

cephalum’s output. It is shown in ﬁgure 5. For the third

and fourth bar of the main riﬀ, guitar and drums parts

were re-composed to match the new tune composed by the

memristors.

Figure 4: The original main riﬀ composed by the

author.

Figure 5: Physarum polycephalum’s rendition of the

main riﬀ for measurement oﬀset of 100%.

4.3 Verse Part 1

The ﬁrst part of the verse essentially contained piano chords,

drums, and vocals. 0% oﬀset subtly altered the harmonic

structure of the verse. 30% output collapsed all t he indi-

vidual notes into one point. This was because the time

b etween the chords was 0, which was obviously not useful

for the composition. 100% made too many variations that

did not suit the genre of the song and it was diﬃcult to

p erceive the harmonic structure. Hence, we chose 0% oﬀset

for this section.

The original chord pattern of the verse was Em—G—

Am—D. All chords were ﬁrst inversions. While processing

the output of Physarum polycephalum, the root notes of

chords were left unchanged. The ﬁrst chord was the same

as the input. The second chord was a triad of the notes

E4, G4, and A4 (the numbers stand for the octave), which

sounded very close to the chord Esus4/G. In order to form a

triad that sounds harmonically more stable, G4 was changed

to C5 and thus, obtaining Esus4/G. The third chord gener-

ated by Physarum polyceph alum contained E2 as the low-

est note, which sounded like the second inversion of Am.

Therefore, the third chord was changed to the second in-

version of Am. The fourth chord generated by Physarum

po lycephalum contained F #4, G4, and A4, which sounded

close to the chord Gsus2/D. However, F #4 and G4 are only

a semi-tone apart and therefore, made the chord sound un-

stable. Hence, F #4 was modiﬁed to D5 and thus obtaining

the chord Gsus2/D.

The original vocal tune composed by the author was a

simple melody line. 0% produced a melody that was similar

to the input, but had irregular rhythms and few variations

in pitches. It did not fulﬁl our goals of making the vocal tune

sound more catchy. 30% produced interesting variations in

the output. However, the output sounded very continuous

to suit a vocal line. Its rhythmic structure did not have

enough stops. 100% oﬀset formed a good foundation for the

vocal line. Furthermore, it went well with the new chords

517

Table 1: Table of tests conduced with PhyBox for diﬀerent values of measurement oﬀset (MO). Numbers

within the MO columns sh ow the total dissimilarity and ‘-’ indicates no test was conducted. The ‘ﬁnal song’

column shows the values of MO chosen for the ﬁnal s ong. The values in bold indicate that they were modiﬁed

to suit the song. There are repetitions of values across rows because the dissimilarities are calculated for

the whole block and not only the section.

Section Physarum MO (%) Final Song

0% 30% 50% 100% 100%R

Main Riﬀ 69 214 - 261 - 100%

Verse Pt. 1 Piano 69 214 - 261 - 0%

Verse Pt. 2 Piano - - - - 150 100%R

Pre-chorus Piano - - - - 150 100%R

Piano Solo 414 738 743 860 313 50% and 100%R

Verse Pt. 1 and Pt. 2 Vocals 106 229 - 265 81 100%

Pre-chorus and Chorus Vocals 106 229 - 265 81 -

comp osed by Physarum polyceph alum. However, few notes

were omitted and altered in order t o suit the song.

4.4 Verse Part 2

The rhythmic structure was unaltered because resistors were

connected instead of memristors for time between note-ons

and duration. During the second bar, the memristors pro-

duced an accidental Bb which was not anticipated by the

comp oser. This accidental (or blue note) is pretty popular

in jazz and blues genres, which gave the song a new ﬂavour.

After performing few modiﬁcations, the arpeggio composed

by Physarum polycephalum was ﬁt into the song.

4.5 Pre-chorus

The pre-chorus of the song mainly consisted of strumming

and piano patterns in close association with the drumbeat

and hence, we wanted to maintain the rhythmic structure.

For this part of the song, we used resistors for rhythmic pa-

rameters and memristors for pitch and velocity. The m em-

ristors changed the voicing of chords and also performed

some chord inversions. Few modiﬁcations were made to in-

corporat e it in the song.

For the vocal tune of the pre-chorus, we felt that the

output of memristors did not match the backing chords.

Furthermore, the original melody line went well with the

subtle changes produced by memristors in t he piano line.

Therefore, we used the original tune.

4.6 Chorus

The chorus of the song had the main riﬀ running in the

background alongside vocals. Therefore, the instrumental

section was a replica of what was done earlier.

In order to compose the vocal tune for the chorus, we

analysed the output of resistors connected to rhythmic pa-

rameters. For the chorus of the song, we felt that the origi-

nal version was more suitable and Physarum polycephalum’s

output did not improve it. As we progressed towards the

chorus, the song was gaining structure and form. Further-

more, it had the main riﬀ of the song playing in the back-

ground. Therefore, there was more rigidity in the number

of suitable musical options. Hence, we chose the original

tune created by the composer.

4.7 Piano Solo

Among all values of measurement oﬀset, 50% was chosen

due to the following explanation. In the original solo com-

p osed by the author, there was a mode shift from E minor

to E Phrygian and then C major. Among all outputs pro-

duced by Physarum polycephalum, 50% created a solo that

was distinctly perceived as G major. Surprisingly, there was

no training data that existed in G major. However, it is the

relative major of the E minor scale. This was a very in-

teresting observation and Physarum polycephalum sounded

like a highly skilled pianist! This demonstrated that the in-

telligence of the organism can be harnessed to create unique

trajectories in creative processes. T here were no modiﬁca-

tions made by the composer on Physarum polycephalum’s

output. Furthermore, 30% oﬀset created few bars that were

p erceived as G major. The output of 30% and 50% were

similar, but 50% was preferred by the composer. 100% oﬀset

created many stops during the solo which made it unsuit-

able. Additionally, 100% oﬀset sounded distinctly diﬀerent

from 30% and 50%.

The ﬁnal piano solo starts with the original one comp osed

by the author. This was mutated with outputs produced

by 0% and 100%R. Just before the scale shift to C major,

Physarum polycephalum’s solo in 50% oﬀset begins, which

lasts for 9 bars. In the ﬁnal song, this part spans from 2

min 22 s to 2 min 42 s.

4.8 Drums

For all values of measurement oﬀset, the musical ﬁles sounded

random and unsuitable for the song. This might be due to

the 4 parameters — pitch, velocity, time b etween note-ons,

and duration are not suitable to analyse percussion tracks.

It might need higher-level musical structures such as time

signatures. Furthermore, in such a song, drums are gen-

erally a supporting instrument. Physarum polycephalum’s

output sounded more like a melody line and not a drum-

b eat. Due to the above observations, we decided to not use

Physarum polycephalum for the drum solo.

4.9 Discussion

This pap er presented the process of composing with Physarum

po lycephalum-based m emristors. As mentioned earlier, the

mo del adopted for composition uses note transitions. There-

fore, it do es not account for higher-level musical structures

like time signatures and harmonic functions. Therefore, we

had to make modiﬁcations in Physarum polycephalum’s out-

put to incorporate it in the song. It is important to note

that modiﬁcations to the output produced by Physarum

po lycephalum was only done in parts where there was a high

involvement of multiple tracks and presence of chords. For

instance, slightly altering piano and vocal lines during the

verse, but not making any modiﬁcations in the main riﬀ.

Furthermore, individual tracks were passed separately into

the PhyBox, which may have contributed to this behaviour.

As no changes were made in outputs produced for the

main riﬀ and piano solo, it suggests that the model can be

improved by further understanding higher-level structures.

Examining how instruments interact with each other opens

up possibilities of creating networks of memristors that col-

518

lectively produce a coherent and ready-to-use output.

The author felt that the memristors have changed the

original composition signiﬁcantly and added its own foot-

print. It was an interactive process where Physarum poly-

cephalum makes suggestions and the composer evaluates

them. However, it is important to not e that many observa-

tions and results were drawn through subjective measures

and aesthetic preferences of the composer.

This paper used diﬀerent values of measurement oﬀset for

diﬀerent sections of the song. In ﬁgure 3, it can be observed

that 0% captures the value at the top of the spike. The next

sample recorded by the PhyBox is after a steep descent.

Therefore, in order to have more ﬂexibility in measurement

oﬀset, the sampling rate of the system needs to be increased.

We used dissimilarity as the metric to compare diﬀer-

ent values of measurement oﬀsets. This can be improved

by borrowing concepts like entropy from information the-

ory [20, 18]. This will enable us t o explore the statistical

structure of music in greater detail.

5. CONCLUSION

This paper presented a composition that was in collabo-

ration with Physarum polycephalum. It showed how the

non-linear behaviour of memristors can be harnessed for the

pro cess of popular music pro duction. It aims to make UC

technologies reach a wider audience so that non-experts in

UC and musicians can adopt them in the process of music

pro duction. Harnessing Physarum polycephalum has great

p otential in computational creativity and popular music.

We showed that resistors can be creatively incorporated

alongside memristors during musical composition. We demon-

strated that a dwell time of 1 s is suitable for obtaining

the memristor’s non-linear behaviour. This is faster than

an earlier model that used 2 s [6]. Dissimilarity values

show that the measurement oﬀset considerably inﬂuenced

the melody produced by the memristor. For each param-

eter, the output created by Physarum polycephalum was

unique. This can be veriﬁed with sound ﬁles provided in

the database.

In this paper, we trained the model with only one song.

Future work would explore how the system can be scaled

up. This might include creating more complicated learn-

ing models such as networks of memristors. Additionally,

higher-level data structures such as time signatures and har-

monic functions need to be explored.

6. ACKNOWLEDGEMENTS

The authors thank Dr. David Moﬀat for proofreading t he

article and oﬀering constructive feedback.

7. REFERENCES

[1] A. Adamatzky. Physarum Machines: Computers from

Slime Mould, volume 74 of A. World Scientiﬁc, 2010.

[2] A. Adamatzky, J. Jones, R. Mayne, S. Tsuda, and

J. Whiting. Logical gates and circuits implemented in

slime mould. In Advances in Physarum Machines,

pages 37–74. Springer, 2016.

[3] O. Bichler, D. Querlioz, S. J. Thorpe, J.-P. Bourgoin,

and C. Gamrat. Extraction of temporally correlated

features from dynamic vision sensors with

spike-timing-dependent plasticity. Neural Networks,

32:339–348, 2012.

[4] M. Bosi and R. E. Goldberg. Introduction to Digital

Audio Coding and Standards. Kluwer Academic

Publishers, 2003.

[5] E. Braund and E. R. Miranda. On building practical

bio computers for real-world applications: Receptacles

for culturing slime mould memristors and component

standardisation. Journal of Bionic Engineering,

14(1):151–162, 2017.

[6] E. Braund, S. Venkatesh, and E. R. Miranda. Phybox:

A programmable interface for physarum

p olycephalum-based memristors. International

Journal of Unconventional Computing,

14(3-4):217–233, 2019.

[7] L. Chua. Memristor-the missing circuit element. IEEE

Transactions on circuit theory, 18(5):507–519, 1971.

[8] S. Colton, G. A. Wiggins, et al. Computational

creativity: The ﬁnal frontier? In Ecai, volume 12,

pages 21–26. Montpelier, 2012.

[9] D. Cope. Experiments in Musical I ntelligence. A-R

Editions, Inc, 1996.

[10] J. D. Fern´andez and F. Vico. Ai methods in

algorithmic composition: A comprehensive survey.

Journal of Artiﬁcial Intelligence Research,

48:513–582, 2013.

[11] E. Gale, A. Adamatzky, and B. de Lacy Costello.

Slime mould memristors. BioNanoScience, 5(1):1–8,

2015.

[12] K. Jones. Comp o sitional applications of stochastic

pro cesses. Computer Music Journal, 5(2):45–61, 1981.

[13] B. Linares-Barranco and T. Serrano-Gotarredona.

Memristance can explain

spike-time-dependent-plasticity in neural synapses.

Nature Precedings, 4, 09 2010.

[14] E. R. Miranda and J. Al Biles. Evolutionary computer

music. S pringer, 2007.

[15] E. R. Miranda, E. Braund, and S. Venkatesh.

Compos ing with biomemristors: Is biocomputing the

new technology of computer music? Computer Music

Journal, 42(3):28–46, 2018.

[16] T. Nakagaki, H. Yamada, and

A. T´oth. Intelligence:

Maze-solving by an amoeboid organism. Nature,

407(6803):470, 2000.

[17] T. Nakagaki, H. Yamada, and A. Toth. Path ﬁnding

by tube morphogenesis in an amoeboid organism.

Biophysica l chemistry, 92(1-2):47–52, 2001.

[18] M. T. Pearce and G . A. Wiggins. Auditory

expect ation: the information dynamics of music

p erception and cognition. Topics in cognitive science,

4(4):625–652, 2012.

[19] S. M. Ross. Stochastic processes. JohnWiley & Sons,

Inc, 2 edition, 1996.

[20] C. E. Shannon. A mathematical theory of

communication. Bell system technical journal,

27(3):379–423, 1948.

[21] G. S. Snider. Spike-timing-dependent learning in

memristive nanodevices. In Proceedings of the 2008

IEEE International Symposium on Nanosca le

Architectures, pages 85–92. IEEE Computer Society,

2008.

[22] D. B. Strukov, G. S. Snider, D. R. Stewart, and R. S.

Williams. The missing memristor found. nature,

453(7191):80, 2008.

[23] P. M. Todd. Computer Music Journal, 13(4):27–43,

1989.

[24] S. Venkatesh, E. Braund, and E. R. Miranda. A

nonlinear approach to generate creative data using

physarum polycephalum-based memristors. In

A. Adamatzky, editor, S lime Mould in Arts and

Architecture. River Publishers, 2019.

519