Title

layout: true
<slide-template title="Applied AI and Data Science on Emergent Technologies" subtitle="Invited Talk VIT Chennai"></slide-template>

---
class: title-slide, center, middle

<title-slide titletype="4th February 2025" title="Applied AI and Data Science on Emergent Technologies" modulenumber="" subtitle="Invited Talk VIT Chennai"></title-slide>

---
# .darkblue[Stations of my] .darkgray[academic] .darkblue[life]

.reference[Three rabbits for illustrating deadlocks instead of dining philosophers problem]

---
# .darkblue[Research Areas]

.more-more-condensed.no-margin[
<table>
<tbody>
<tr>
<td style="width:48%;vertical-align:top">.more-more-condensed[
- .darkblue[Artificial Intelligence, Machine Learning and Data Science]
  - LLMs, Agentic Workflows, Mathematical Optimizations, Graph Neural Networks, Chatbots, Reasoning
- .darkblue[Data Management Tasks]
  - Query Processing & Opt., Indexing, Mapping, Compression, Replication, Caching, Transaction Handling
- .darkblue[Data Models]
  - Knowledge Graphs, Semantic Web, Property Graphs, Relational Data, XML
- .darkblue[Types of Data]
  - Big Data, Data Streams
]
</td>
<td style="vertical-align:top">.more-more-condensed[
- .darkblue[Emergent Hardware Technologies]
  - Many-Core CPU, GPU, FPGA, Quantum Computer
- .darkblue[Platforms]
  - Internet, Internet of Things, Cloud, Post-Cloud (Fog/Edge/Dew Computing), P2P, Mobile, Parallel and Main Memory Servers
- .darkblue[Advanced Applications]
  - Citizen Science, Customer Communications, Pandemics like Covid-19, Software Vulnerability Prediction
- .darkblue[Sustainability]
  - Sustainable Computing/AI, Applications for Sustainability
]
</td>
</tr>
</tbody>
</table>
]

---
# .darkblue[Lectures] by Sven Groppe

.more-more-condensed.no-margin.no-padding[
.darkblue[50 lectures] in the areas of
- .darkblue[Information Systems] .filledCircle.darkgreenbackground[N] .filledCircle.darkgreenbackground[T] (inclusive Knowledge Graphs, LLMs and Graph Neural Networks) .darkgray[(since 2024)]
- .darkblue[Quantum Computing] .filledCircle.darkgreenbackground[N] .filledCircle.darkgreenbackground[T] (inclusive Quantum Machine Learning) .darkgray[(since 2022)]
- .darkblue[Semantic Web] .filledCircle.darkgreenbackground[N] .filledCircle.darkgreenbackground[T] .darkgray[(since 2011)]
- .darkblue[Databases] .darkgray[(2014, since 2024)]
- .darkblue[Mobile and Distributed Databases] .filledCircle.darkgreenbackground[N] .darkgray[(since 2008)]
- .darkblue[Cloud and Web Technologies] .filledCircle.darkgreenbackground[N] .filledCircle.darkgreenbackground[T] .darkgray[(since 2015)]
- .darkblue[Algorithms and Datastructures] .darkgray[(2014)]
- .darkblue[XML Databases] .filledCircle.darkgreenbackground[N] .darkgray[(2007)]
- .darkblue[Compiler Design] .filledCircle.darkgreenbackground[N] .filledCircle.darkgreenbackground[T] .darkgray[(2006/2007)]
- .darkblue[Next Web Generation] .darkgray[(2006, together with M. Zaremba)]

<div class="margin-bottom"></div>
.small-padding.legend.small-font[.filledCircle.darkgreenbackground[N] Newly designed .filledCircle.darkgreenbackground[T] Online Tutorials]
]

---
# .darkblue[Supervision and Publication Record]

.reference[<sup>1</sup> according to <a href="http://www.conferenceranks.com/" target="_blank">Conference Ranks</a>]

.more-more-condensed[
- .darkblue[3 supervised dissertations], 6 current PhD students
- .darkblue[$\approx$ 100 bachelor/master]/diploma .darkblue[thesis]/student projects
- .darkblue[$>$ 191 publications]
  - .darkblue[16 publications at A/A1 ranked conferences]<sup>1</sup>
  - .darkblue[195 co-authors] affiliated with organizations .darkblue[in 28 countries on 6 continents]<table><tr><td style="width:50%"><object id="map_co-authors" type="image/svg+xml" data="https://www.ifis.uni-luebeck.de/~groppe/resources/images/coauthorMap.svg" class="fullwidth"><img alt="World map co-authors of Sven Groppe" src="https://www.ifis.uni-luebeck.de/~groppe/resources/images/coauthorMap.svg" class="fullwidth"></object></td><td style="width:50%"></td></tr></table>
  - .darkblue[48] publications (25%) are .darkblue[co-authored by] bachelor/master .darkblue[students] (being students at time of writing)
]

---
# .darkblue[Project Grants ($\approx$ 2M <span class="material-icons"></span>)]

.more-more-condensed.small-font24.no-margin.no-padding[
<ol>
<li> .filledCircle.darkredbackground[D] .filledCircle.darkredbackground[I] .darkblue[High Quality Knowledge Graphs from recent English, French and German Emergent Trends with the example of COVID-19] .darkgray[(2022-2025)]</li>
<li> .filledCircle.darkredbackground[B] .darkblue[QC4DB: Accelerating Relational Database Management Systems via Quantum Computing] .darkgray[(2022-2025)]</li>
<li> .filledCircle.darkredbackground[D] .darkblue[Hybrid<sup>2</sup>-Indexstrukturen für Hauptspeicherdatenbanken] .darkgray[(2019-2024)]</li>
<li> .filledCircle.darkredbackground[D] .darkblue[BigSIoT: Big Data Management for the Semantic Internet of Things] .darkgray[(2020-2023)]</li>
<li> .filledCircle.darkredbackground[D] .darkblue[Hardwarebeschleunigung von Semantic Web Datenbanken durch dynamisch rekonfigurierbare FPGAs] .darkgray[(2013-2015)]</li>
<li> .filledCircle.darkredbackground[W] .darkblue[Beschleunigung relationaler Datenbanken mittels laufzeitadaptiver FPGA-Cluster] .darkgray[(2013-2015)]</li>
<li> .filledCircle.darkredbackground[D] .darkblue[Logisch und Physikalisch Optimierte Semantic Web Datenbank-Engine] .darkgray[(2007-2009)]</li>
</ol>
<div class="margin-bottom"></div>
.small-padding.legend.small-font[.filledCircle.darkredbackground[D] DFG projects .filledCircle.darkredbackground[I] International Project .filledCircle.darkredbackground[W] BMWi/ZIM .filledCircle.darkredbackground[B] BMBF]
]

---
# .darkblue[Scientific Services]

.reference[Detailed Information: <a target="_blank" href="https://www.ifis.uni-luebeck.de/~groppe" style="color:white" class="bold">https://www.ifis.uni-luebeck.de/~groppe</a>]

.more-more-condensed[
- <span class="darkblue">General Chair</span>
  <ul>
  <li><span class="darkblue">The International Conference on Applied Machine Learning and Data Analytics (AMLDA)</span> '23</li>
  <li><span class="darkblue">International Semantic Intelligence Conference (ISIC)</span> ('21-'22)</li>
  <li><span class="darkblue">International Health Informatics Conference (IHIC)</span> ('22-'24)</li>
  <li><span class="darkblue">International EdTech Conference (IEdTC)</span> '23</li>
  </ul>
- <span class="darkblue">Workshop</span> Chairs
  <ul>
      <li><span class="darkblue">Quantum Data Science and Management (QDSM)</span>@VLDB ('23-'24)</li>
      <li><span class="darkblue">Semantic Big Data (SBD)</span>@SIGMOD ('16-'20)</li>
	  <li><span class="darkblue">Big Data in Emergent Distributed Environments (BiDEDE)</span>@SIGMOD ('21-'23)</li>
	  <li><span class="darkblue">Very Large Internet of Things (VLIoT)</span>@VLDB ('17-'22)</li>
  </ul>
- many other <span class="darkblue">scientific services</span>
  <ul><li>$\approx$ 128 <span class="darkblue">PC memberships<span></li>
	  <li><span class="darkblue">reviewer</span> of $\approx$ 42 journals</li>
	  <li><span class="darkblue">editor</span> of 4 journals</li>
	  <li>...</li></ul>
]

---
# .darkblue[Interplay of LLMs and KGs]

.reference[<a href="https://vldb.org/workshops/2024/proceedings/LLM+KG/LLM+KG-9.pdf" target="_blank">Khorashadizadeh et al., Research Trends for the Interplay between Large Language Models and Knowledge Graphs, LLM+KG@VLDB'24</a>, <a href="https://doi.org/10.48550/arxiv.2406.08223" target="_blank">ArXiv version</a>]

---
# .darkblue[KG Construction with LLMs]<br/>.darkgray[(Project with Uni Paris/Toulouse)]

.reference[<sup>*</sup> <a href="https://doi.org/10.48550/arXiv.2305.08804" target="_blank">H.Khorashadizadeh et al., Exploring In-Context Learning Capabilities of Foundation Models for Generating KGs from Text, Text2KG@ESWC 2023</a>]

.more-more-condensed.no-margin[
<div class="textcenter"><img src="../../img/keynotes/bv/KGExtractionFromLLM.svg" width="100%"/></div>
- .darkblue[Knowledge graph (KG) construction, completion, fact validation, extraction facts from unseen text]<sup>*</sup> (limitations due to hallucination)
]

---
# .darkblue[KG Construction with LLMs]<br/>.darkgray[(Project with Uni Paris/Toulouse)]

.reference[<a href="https://doi.org/10.1007/978-3-031-70239-6_31" target="_blank">Ezzabady et al., Towards Generating High-Quality Knowledge Graphs by Leveraging Large Language Models, NLDB 2024</a>]

.more-more-condensed.no-margin[
<div class="textcenter"><img src="../../img/keynotes/bv/KGConstructionWithLLMs.svg" style="width:80%"/></div>
<ul><li>.darkblue[Result] (comparison to manual labeling)<div class="margin-bottom"><table class="result small-font tablesmallpadding">
<thead>
<tr class="header">
<th style="text-align: center;"></th>
<th style="text-align: center;"><span class="smallcaps">Cov</span></th>
<th style="text-align: center;"><span class="smallcaps">Gen</span></th>
<th style="text-align: center;"><span class="smallcaps">Db</span></th>
</tr>
</thead>
<tbody>
<tr class="odd">
<td style="text-align: center;"><strong>Macro F1-score</strong></td>
<td style="text-align: center;">66.60</td>
<td style="text-align: center;"><strong>76.93</strong></td>
<td style="text-align: center;">72.85</td>
</tr>
</tbody>
</table></div>
<ul><li><span class="darkblue">COV:</span> Represents the original triples extracted <span class="darkblue">using GPT-4</span></li>
<li><span class="darkblue">GEN:</span> COV with <span class="darkblue">augmented data generated by GPT-4</span></li>
<li><span class="darkblue">DB:</span> Merges the original triples from <span class="darkblue">COV with triples from an external data source</span> (with daily statistical updates and policy actions)</li></ul>
</li></ul>
]

---
# .darkblue[KG Construction with LLMs]<br/>.darkgray[(Project with Uni Paris/Toulouse)]

.reference[Khorashadizadeh et al. Construction and Canonicalization of Economic Knowledge Graphs with LLMs, KGSWC 2024]

<table>
<tbody>
<tr>
<td style="width:40%">
.more-more-condensed.no-margin[
- Approach to .bold.darkblue[Open Information Extraction]
  - extracting structured information directly from unstructured text <span class="bold darkblue">without a predefined schema</span>
]
</td>
<td style="width:60%">
<div class="textcenter more-more-condensed bold">
.darkblue[Pipeline:]<br/>
<img src="../../img/keynotes/covid-19/OpenIE_Pipeline.svg" align="top" class="fullwidth"></img>
</div>
</td>
</tbody>
</table>

---
# .darkblue[Knowlege Graph for Analysis of COVID-19 Policies] .darkgray.small-font[(Policy Importance in 183 Nations)]

.reference[Khorashadizadeh et al., Assessing the Impact of Government Policies on COVID-19 Spread: A Machine Learning Approach, LOD2024]

.more-more-condensed[
<div class="textcenter">
<img src="../../img/keynotes/bv/total_coef_RF_new_cases_per_million.svg" align="top" style="width:95%"></img>
</div>
]

---
# .darkblue[Knowlege Graph for Analysis of COVID-19 Policies] .darkgray.small-font[(Policy Importance in 183 Nations)]

.reference[Khorashadizadeh et al., Assessing the Impact of Government Policies on COVID-19 Spread: A Machine Learning Approach, LOD2024]

.more-more-condensed[
<div class="textcenter">
<img src="../../img/keynotes/bv/total_coef_RF_new_deaths_per_million.svg" align="top" style="width:95%"></img>
</div>
]

---
# .darkblue[Nations clustered according to COVID-19 Policies Importance] .darkgray.small-font[(Policy Importance in 183 Nations)]

.reference[Khorashadizadeh et al., Assessing the Impact of Government Policies on COVID-19 Spread: A Machine Learning Approach, LOD2024]

.more-more-condensed[
<div class="textcenter">
<img src="../../img/keynotes/bv/WorldClusteredByPolicyImpact.svg" align="top" style="width:100%"></img>
</div>
]

---
# .darkblue[LLMs in Citizen Science Platforms]<br/>.small-font26[(Les Herbonautes, project with MNHN/Uni Paris)]

.reference[<a href="https://doi.org/10.3390/computers14010021" target="_blank">Kessel et al., Impact of Chatbots on User Experience and Data Quality on Citizen Science Platforms, Computers 14(1), 2024</a>]

.more-more-condensed.small-font24[
- launched by .darkblue[Muséum national d'Histoire naturelle (MNHN)]/Paris
- collaborative way of contributing to the .darkblue[creation of a scientific database based on] the .darkblue[millions of photos of plants] in the national network of naturalist collections (Récolnat)
<div class="textcenter">
<img src="../../img/keynotes/bv/LesHerbonautes.jpg" align="top" style="width:75%" class="margin-bottom"></img>
</div>
]

---
# .darkblue[LLMs in Citizen Science Platforms]

<table class="padding-top" style="border-collapse:collapse">
<tr>
<td style="width:70%;vertical-align:top;">
.more-more-condensed.no-margin[
- LLM in form of a network of (LLM, human, code) agents as basis for an .darkblue[easy-to-use chatbot]
  - .darkblue[for guiding] users .darkblue[through] the .darkblue[data validation] process and .darkblue[with which users can] retrieve important information by .darkblue[posing related questions] for adding missing information (as some kind of easy-to-use tool for their research) 
  - .darkblue[for high-level analysis] with visual presentation of the analysis results in the form of maps, charts and/or in text
]
</td>
<td style="width:30%;vertical-align:top;">
<img src="../../img/keynotes/bv/ConversibleAgents.svg" class="fullwidth"></img>
</td>
</tr>
<tr>
<td colspan="2">
.more-more-condensed.no-margin[
<ul><ul><li>.darkblue[for detecting anomalies and marking any wrong data], or data and results of the chatbot they do not trust, as a trigger for data validation</li>
<li>.darkblue[for multi-lingual support]</li>
</ul></ul>
]
</td>
</tr>
</table>

---
# .darkblue[Multi-Agent Network in Citizen Science

<div class="textcenter">
<img src="../../img/keynotes/bv/ArchitectureLesHerbonautsV3.png" align="top" style="width:70%"></img>
</div>
]

---
# .darkblue[Multi-Agent Network in Citizen Science

<div class="textcenter">
<img src="../../img/keynotes/bv/LesHerbonautesChatBot.png" align="top" style="width:70%"></img>
</div>
]

---
# .darkblue[AI in Cyber Security<br/><span class="small-font32">Graph Neural Networks for Softw. Vulnerability Detection</span>] 
.reference[J. Groppe, S. Groppe, R. Möller, Variables are a Curse in Software Vulnerability Prediction. DEXA'23]

.more-more-condensed[
- Input: Software code
- Output: Is the code vulnerable? 
  - Future work: "Automatic" patch generation
<img src="../../img/keynotes/bv/ASG.svg" style="width:72%"/>
]

---
# .darkblue[AI in Cyber Security<br/><span class="small-font32">Graph Neural Networks for Softw. Vulnerability Detection</span>]
.reference[J. Groppe, S. Groppe, R. Möller, Variables are a Curse in Software Vulnerability Prediction. DEXA'23]
  
.more-more-condensed[
<table>
<tr>
<td>

3-properties encoding<br/>per AST node<br/>(with variable names)
<table class="result tablesmallpadding small-font">
<tr>
<th>
Construct
</th>
<th>
Class
</th>
<th>
Name
</th>
<th>
Type
</th>
</tr>
<tr>
<th>
int vname
</th>
<td>
varDecl
</td>
<td>
<span style="color:red">vname</span>
</td>
<td>
int
</td>
</tr>
<tr>
<th>
char[6] cname
</th>
<td>
varDecl
</td>
<td>
<span style="color:red">cname</span>
</td>
<td>
char[<span style="color:red">6</span>]
</td>
</tr>
<tr>
<th>
if
</th>
<td>
control
</td>
<td>
if
</td>
<td>
-
</td>
</tr>
<tr>
<th>
0.05
</th>
<td>
literal
</td>
<td>
<span style="color:red">0.05</span>
</td>
<td>
float
</td>
</tr>
<tr>
<th>
*
</th>
<td>
mathOp
</td>
<td>
mul
</td>
<td>
-
</td>
</tr>
<tr>
<th>
fputs(...)
</th>
<td>
call
</td>
<td>
fputs
</td>
<td>
-
</td>
</tr>
<tr>
<th>
vname
</th>
<td>
ident
</td>
<td>
<span style="color:red">vname</span>
</td>
<td>
int
</td>
</tr>
<tr>
<th>
stdout
</th>
<td>
ident
</td>
<td>
stdout
</td>
<td>
-
</td>
</tr>
<tr>
<th>
{...}
</th>
<td>
block
</td>
<td>
-
</td>
<td>
-
</td>
</tr>
</table>

</td>
<td style="padding-left:1em">

3-properties encoding<br/>per AST node<br/>(without variable names)
<table class="result tablesmallpadding small-font">
<tr>
<th>
Construct
</th>
<th>
Class
</th>
<th>
Name
</th>
<th>
Type
</th>
</tr>
<tr>
<th>
int vname
</th>
<td>
varDecl
</td>
<td>
-
</td>
<td>
int
</td>
</tr>
<tr>
<th>
char[6] cname
</th>
<td>
varDecl
</td>
<td>
-
</td>
<td>
char[N]
</td>
</tr>
<tr>
<th>
if
</th>
<td>
control
</td>
<td>
if
</td>
<td>
-
</td>
</tr>
<tr>
<th>
0.05
</th>
<td>
literal
</td>
<td>
-
</td>
<td>
float
</td>
</tr>
<tr>
<th>
*
</th>
<td>
mathOp
</td>
<td>
mul
</td>
<td>
-
</td>
</tr>
<tr>
<th>
fputs(...)
</th>
<td>
call
</td>
<td>
fputs
</td>
<td>
-
</td>
</tr>
<tr>
<th>
vname
</th>
<td>
ident
</td>
<td>
var
</td>
<td>
int
</td>
</tr>
<tr>
<th>
stdout
</th>
<td>
ident
</td>
<td>
stdout
</td>
<td>
-
</td>
</tr>
<tr>
<th>
{...}
</th>
<td>
block
</td>
<td>
-
</td>
<td>
-
</td>
</tr>
</table>

</td>
</tr>
</table>
]

---
# .darkblue[AI in Cyber Security<br/><span class="small-font32">Graph Neural Networks for Softw. Vulnerability Detection</span>]

.reference[J. Groppe, S. Groppe, R. Möller, Variables are a Curse in Software Vulnerability Prediction. DEXA'23]

.more-more-condensed[
- Evaluation
  - Improving accuracy
<table class="result tablesmallpadding smaller-font">
    <tr>
        <th></th>
        <th></th>
        <th></th>
        <th colspan="2">Chromium+Debian</th>
        <th colspan="2">FFmpeg+Quemu</th>
        <th colspan="2">VDISC</th>
    </tr>
    <tr>
        <th>Model</th>
        <th>Graph</th>
        <th>Encoding</th>
        <th>Acc</th>
        <th>F1</th>
        <th>Acc</th>
        <th>F1</th>
        <th>Acc</th>
        <th>F1</th>
    </tr>
    <tr>
        <td>codeAST</td>
        <td>AST</td>
        <td>code</td>
        <td>92.01</td>
        <td>30.20</td>
        <td>55.36</td>
        <td>57.01</td>
        <td>77.82</td>
        <td>75.57</td>
    </tr>
    <tr>
        <td>3propASG</td>
        <td>ASG</td>
        <td>3-Prop.</td>
        <td class="bold">92.34</td>
        <td class="bold">44.97</td>
        <td class="bold">60.35</td>
        <td>62.30</td>
        <td class="bold">81.27</td>
        <td class="bold">79.86</td>
    </tr>
    <tr>
        <td>codeAST+</td>
        <td>AST+</td>
        <td>code</td>
        <td>90.89</td>
        <td>25.86</td>
        <td>58.38</td>
        <td>46.66</td>
        <td>75.67</td>
        <td>74.49</td>
    </tr>
    <tr>
        <td>3propASG+</td>
        <td>ASG+</td>
        <td>3-Prop.</td>
        <td class="bold">92.34</td>
        <td>44.59</td>
        <td>57.04</td>
        <td class="bold">62.99</td>
        <td>80.94</td>
        <td>79.63</td>
    </tr>
</table>  
  - Improving memory footprint
  <table class="result tablesmallpadding smaller-font">
    <tr>
        <th>#nodes</th>
        <th>#tokens</th>
        <th>memory code-based</th>
        <th>memory 3-prop.</th>
        <th>code-based/3-prop.</th>
    </tr>
    <tr>
        <td>4,409</td>
        <td>33,659</td>
        <td>59G</td>
        <td class="bold">5.3M</td>
        <td>11,220</td>
    </tr>
    <tr>
        <td>7,012</td>
        <td>54,157</td>
        <td>152G</td>
        <td class="bold">8.4M</td>
        <td>18,052</td>
    </tr>
    <tr>
        <td>12,077</td>
        <td>96,805</td>
        <td>468G</td>
        <td class="bold">14.5M</td>
        <td>32,268</td>
    </tr>
</table>
  - Reducing vocabularity explosion
  <table class="result tablesmallpadding smaller-font">
  <tr>
  <th>Dataset</th>
  <th>Code-Based</th>
  <th>3-prop.</th>
  <th>3-prop./Code-Based</th>
  </tr>
  <tr>
  <td>Chromium+Debian</td>
  <td>57,027</td>
  <td>35,416</td>
  <td>62.10%</td>
  </tr>
  <tr>
  <td>FFmpeg+Qume</td>
  <td>66,791</td>
  <td>45,795</td>
  <td>68.56%</td>
  </tr>
  <tr>
  <td>VDSIC</td>
  <td>449,148</td>
  <td>312,948</td>
  <td>69.68%</td>
  </tr>
  </table>
]

---
# .darkblue[AI in Customer Communication]<br/>.small-font26[(Project with ZVO/Germany)]

.reference[J. Groppe, R. Schlichting, S. Groppe, R. Möller, Deep Learning-based Classification of Customer Communications of a German Utility Company, ISIC'22]

.more-more-condensed[
- .darkblue[Real-world] data .darkblue[is heterogeneous]
<table class="fullwidth smaller-font">
<tr>
<td style="width:15%;text-align:end">
#
</td>
<td class="textcenter" style="width:70%">
<div class="margin-bottom">
<img src="../../img/keynotes/bv/ZVO_comm_length.png" align="top" class="fullwidth"></img><br/>length of communication
</div>
</td>
<td style="width:15%">
</td>
</tr>
<tr colspan="2">

<div class="margin-bottom">
<table class="result tablesmallpadding smaller-font" style="margin-left:auto;margin-right:auto">
<tr>
<th>#classes</th><td>1</td><td>2</td><td>3</td><td>4</td><td>5</td><td>6</td><td>7</td><td>8</td><td>9</td><td>10</td><td>11</td><td>12</td><td>13</td><td>14</td><td>15</td>
</tr>
<tr>
<th>#samples</th><td>119655</td><td>12031</td><td>1079</td><td>199</td><td>31</td><td>16</td><td>6</td><td>7</td><td>2</td><td>3</td><td>5</td><td>3</td><td>2</td><td>2</td><td>4</td>
</tr>
</table>
</div>

</tr>
<tr cospan="2">

<table class="result tablesmallpadding smaller-little-font" style="margin-left:auto;margin-right:auto">
<tr>
<th>Class</th><td>C0</td><td>C1</td><td>C2</td><td>C3</td><td>C4</td><td>C5</td><td>C6</td><td>C7</td><td>C8</td><td>C9</td><td>C10</td><td>C11</td><td>C12</td><td>C13</td><td>C14</td><td>C15</td><td>C16</td><td>C17</td>
</tr>
<tr><th>#Comm</th><td>9770</td><td>21062</td><td>14245</td><td>235</td><td>1307</td><td>251</td><td>4009</td><td>3610</td><td>9410</td><td>23533</td><td>20676</td><td>43</td><td>2866</td><td>9617</td><td>8144</td><td>12126</td><td>4748</td><td>2700</td>
</tr> 
</table>

</tr>
</table>
]

---
# .darkblue[AI in Customer Communication]<br/>.small-font26[(Project with ZVO/Germany)]

.reference[J. Groppe, R. Schlichting, S. Groppe, R. Möller, Deep Learning-based Classification of Customer Communications of a German Utility Company, ISIC'22]

.more-more-condensed[
- .darkblue[Performance] of various approaches

<table class="result tablesmallpadding smaller-little-font" style="margin-left:auto;margin-right:auto">
<tr>
<th>Model</th><th>MLP-TF</th><th>MLP-TFIDF</th><th>CNN-sl</th><th>CNN-G</th><th>CNN-G-u</th><th>CNN-F</th><th>CNN-F-u</th><th>LSTM-sl</th><th>LSTM-G</th><th>LSTM-G-u</th><th>LSTM-F</th><th>LSTM-F-u</th>
</tr>
<tr>
<th>Accuracy</th><td>55.1</td><td>53.2</td><td>58.7</td><td>50.6</td><td>57.5</td><td>55.1</td><td>56.4</td><td>57,4</td><td>53.5</td><td>61.2</td><td>51.2</td><td>58.0</td>
</tr>
<tr>
<th>Precision</th><td>61.6</td><td>59.2</td><td>66.3</td><td>57.2</td><td>64.8</td><td>62.4</td><td>64.4</td><td>64.4</td><td>60.0</td><td>69.0</td><td>57.3</td><td>65.2</td>
</tr>
<tr>
<th>Recall</th><td>60.4</td><td>58.2</td><td>65.8</td><td>56.4</td><td>64.1</td><td>62.1</td><td>64.3</td><td>62.6</td><td>58.3</td><td>67.9</td><td>55.6</td><td>63.7</td>
</tr>
<tr>
<th>AUC</th><td>79.7</td><td>78.7</td><td>82.2</td><td>77.6</td><td>81.5</td><td>80.4</td><td>81.5</td><td>80.6</td><td>78.5</td><td>83.2</td><td>77.2</td><td>81.1</td>
</tr>
<tr>
<th>Time (s)</th><td>961</td><td>285</td><td>1388</td><td>1805</td><td>1674</td><td>2282</td><td>1005</td><td>5728</td><td>14817</td><td>7973</td><td>22757</td><td>5775</td>
</tr>
</table>

<ul><ul><li>.darkgray[BERT-transfomer even worse]</li></ul></ul>
]

---
# .darkblue[AI in Customer Communication]<br/>.small-font26[(Project with ZVO/Germany)]

.reference[J. Groppe, R. Schlichting, S. Groppe, R. Möller, Deep Learning-based Classification of Customer Communications of a German Utility Company, ISIC'22]

.more-more-condensed[
- .darkblue[Visualization] of all communications
  - .darkblue[Similar communications are close] to each other
  - .darkblue[Communications of one class] are marked in red
     - <span class="darkblue">not close</span> to each other, difficult to group them<br/><span class="darkblue">$\Rightarrow$ Low separability of classes</span>
	 
<div class="textcenter">
<img src="../../img/keynotes/bv/ZVO_umap_RNNgu_1d_0.jpg" align="top" width="30%"></img>
<img src="../../img/keynotes/bv/ZVO_umap_RNNgu_1d_1.jpg" align="top" width="30%"></img>
<img src="../../img/keynotes/bv/ZVO_umap_RNNgu_1d_2.jpg" align="top" width="30%"></img><br/>...
</div>
]

---
# .darkblue[Green Computing in Industry 4.0]<br/>.small-font26[(Project with Bosch)]

.reference[S. Paasche, S. Groppe, GreenCC: A Hybrid Approach to Sustainably Validate Manufacturing Data in Industry 4.0 Environments, DATA'23]

.more-more-condensed[
<div class="textcenter">
<img src="../../img/keynotes/bv/SustainableCC.svg" align="top" width="73%"></img>
</div>
<ul>
<li>.darkblue[Energy savings by lightweight components] during normal operation and switching on full components for inconsistency handling
<ul><li>.darkblue[CO2e emissions can be reduced by a factor of] about .darkblue[0.6]</li><li>in one year 262 kgCO2e in EU for a medium-sized plant</li></ul></li>
</ul>
]

---
# .darkblue[Hybrid Multi-Model Multi-Platform (HM3P) Database]

.reference[S. Groppe, J. Groppe, Hybrid Multi-Model Multi-Platform (HM3P) Databases, DATA 2020.]

.more-condensed[
<img src="../../img/keynotes/HM3PDatabase.svg" align="top" width="100%"></img>
<span class="darkgreenbackground filledCircle">+</span> full and uniform .darkgreen[data integration] at database level<br/>
<span class="darkgreenbackground filledCircle">+</span> .darkgreen[performance]: fully optimized across different data models<br/>
<span class="darkgreenbackground filledCircle">+</span> transparent .darkgreen[fault-tolerance]<br/>
<span class="darkgreenbackground filledCircle">+</span> SQL .darkgreen[standards]: .small-font22[relational ('87), XML ('03), temporal ('11), JSON ('16), Multi-dimensional Arrays ('19), schemaless ('19), streams ('20?), property graphs ('21?)]<br/>
<span class="darkgreenbackground filledCircle">+</span> <span class="bold">.darkgreen[features of different] types of .darkgreen[databases running on different platforms] can be used
</span>]

---
# Variant: .darkblue[Semantic] HM3P (.darkblue[S]HM3P) DB

.reference[S. Groppe, Semantic Hybrid Multi-Model Multi-Platform (SHM3P) Databases, ISIC 2021.]

.more-condensed.no-padding.no-margin[
<img src="../../img/keynotes/SHM3P.svg" align="top" width="100%"></img>
<ul>
<li><span class="darkblue">Semantic Layer as glue</span> between other models and platforms</li>
<li><span class="darkblue">new challenges</span> like integrating different types of reasoners in a transparent global reasoner</li>
</ul>
<table>
<tbody>
<tr>
<td style="width:55%;vertical-align:top">
<span class="darkgreenbackground filledCircle">+</span> .darkgreen[Features of HM3P] databases<br/>
<span class="darkgreenbackground filledCircle">+</span> Easier .darkgreen[data integration]<br/>
</td>
<td style="width:45%;vertical-align:top">
<span class="darkredbackground filledCircle">-</span> .darkblue[Performance issues] may occur due to semantic layer
</td>
</tr>
</tbody>
</table>
]

---
# Semantic Web DBMS .darkblue[LUPOSDATE]

.reference[S. Groppe:Data Management and Query Processing in Semantic Web Databases,Springer,2011. Open Source:<a href="https://github.com/luposdate/luposdate" target="_blank">https://github.com/luposdate/luposdate</a>]

---
# The Power of .darkblue[Multi-Platform]: .darkblue[LUPOSDATE3000]

.more-more-condensed.no-margin[
<ul><li><span class="darkblue">LUPOSDATE300: Open Source Multi-Platform SW database</span>
<ul><li>Design Choice: Implemented in Kotlin for multi-platform support</li></ul></li></ul>
<table>
<tbody>
<tr>
<td class="halfwidth" style="vertical-align:top">
.more-more-condensed[
<ul><ul><li>.darkblue[ultra-fast in jvm]...
  <div class="textcenter" style="line-height:0.6;"><img src="../../img/keynotes/LUPOSDATE3000Experiments.svg" align="top" class="fullwidth"></img><br/>.smaller-font[<a href="https://doi.org/10.18420/btw2021-12" target="_blank">B. Warnke, M.W. Rehan, S. Fischer, S. Groppe: Flexible data partitioning schemes for parallel merge joins in semantic web queries in: BTW'21</a>]</div></li></ul>
]
</td>
<td class="halfwidth" style="vertical-align:top">
.more-more-condensed[
<ul><ul><li>...but also .darkblue[enabling web demos running completely in] the .darkblue[browser!]
    <div class="textcenter" style="line-height:0.6;"><video controls="controls" preload="preload" class="fullwidth">
  <source src="https://www.ifis.uni-luebeck.de/~groppe/soundofdatabases/resources/CommonSubexpressions.mp4">
  Your browser does not support the video element!
</video><br/><span class="smaller-font"><a href="http://www.vldb.org/pvldb/vol14/p2695-groppe.pdf" target="_blank">S. Groppe, R. Klinckenberg, B. Warnke. Sound of Databases: Sonification of a Semantic Web Database Engine. PVLDB, 14(12), 2021</a></span></div></li></ul>
]
</td>
</tr>
</tbody>
</table>
]

---
# .darkblue[Reinforcement Learning] for .darkblue[Query Optimization]

.reference[Maja et al., Hype or Heuristic? Quantum Reinforcement Learning for Join Order Optimisation, arXiv 2405.07770]

---
# .darkblue[Reinforcement Learning] for .darkblue[Query Opt.]

.reference[Warnke, Martens, Winker, Groppe, Adhiyaman, Srinivasan, Shridevi, ReJOOSp: Reinforcement learning for join order optimization in SPARQL, BDCC, 8(7)]

---
# Execution Times of .darkblue[Query Optimization]

.reference[Warnke, Martens, Winker, Groppe, Adhiyaman, Srinivasan, Shridevi, ReJOOSp: Reinforcement learning for join order optimization in SPARQL, BDCC, 8(7)]

<div class="textcenter no-margin">
<img src="../../img/keynotes/bv/join_ordering_dp_ml_greedy_times.svg" align="top" style="width:77%"></img>
<br/><span class="smaller-font">ML model: maskable Proximal Policy Optimization (PPO) model(25) of stable baselines3 1.5.0, Python 3.9.7</span>
</div>

---
# .darkblue[Query Opt.] via .darkblue[Reinforcement Learning]

.reference[<a href="https://doi.org/10.3390/bdcc8070071" target="_blank">Warnke, Martens, Winker, Groppe, Adhiyaman, Srinivasan, Shridevi, ReJOOSp: Reinforcement learning for join order optimization in SPARQL, BDCC, 8(7)</a>]

---
# .darkblue[Query Opt. & Routing in IoT]

.reference[<a href="https://doi.org/10.1145/3589462.3589497" target="_blank">Warnke, Fischer, Groppe. ReJOOSp: Distributed SPARQL Queries in Collaboration with the Routing Protocol, IDEAS 2023</a>]

.more-more-condensed.no-margin[
- Design Choice: .darkblue[Combining Routing and Query Optimization] in IoT
  - Q9 with 11 triple patterns: <span class="darkblue">Network traffic reduces by</span>
     - <span class="darkblue">routing-assisted join order</span> (instead of centralized query optimization)
	 - <span class="darkblue">dynamic relocation of operators</span> whenever the result of a join is more significant than its inputs (4% for large and 23% for small data scales)
]
<div class="textcenter no-margin">
<img src="../../img/keynotes/IoT/IoTRoutingOptQ9.svg" align="top" width="65%"></img>
</div>
<div class="smaller-font no-margin darkgray" style="line-height:0.9">Transferred data for benchmark queries in percent (smaller $\rightsquigarrow$ better) dependent on option. More significant effects are in the center of the circle.</div>

---
# Architectures of .darkblue[Emergent Hardware]

.reference[Extended from <a target="_blank" href="http://www.imprs-dynamics.mpg.de/pdfs/Plessl_talk.pdf">C. Plessl, Accelerating Scientific Computing with Massively Parallel Computer Architectures, IMPRS Winter School, Wroclaw, 2012</a>]

---
# <table>
<tbody>
<tr>
<td width="25%"><img src="../../img/keynotes/FPGAArchitecture.svg" align="top" width="100%"></img></td>
<td>.darkblue[LUPOSDATE on FPGA]</td>
</tr>
</tbody>
</table>

.reference[S. Werner, Hybrid Architecture for Hardware-accelerated Query Proc. in Semantic Web DB based on Runtime Reconfigurable FPGAs, 2017]

<div class="more-more-condensed no-margin">
<table class="fullwidth">
<tr>
<td style="width:40%">
<ul class="no-margin"><li>Design Choice:<ul><li>At system deployment time generated <span class="darkblue">RMs are configured into RPs</span> at system runtime<br/><span class="darkblue">$\rightsquigarrow$ avoids long synthesis time</span></li></ul></li></ul>
</td>
<td style="width:20%"><img src="../../img/keynotes/EmptySemiStaticOperatorgraph.svg" align="top" width="100%"></img></td>
<td style="width:20%" class="textcenter">.small-font[SP<sup>2</sup>B Query 4]<br/><span class="material-icons"></span><br/>
.smaller-font.darkgray[RM: Reconfigurable Module<br/>RP: Reconfigurable Partition<br/>
SRE: Semi-static Routing Element]</td>
<td style="width:20%"><img src="../../img/keynotes/FullSemiStaticOperatorgraph.svg" align="top" width="100%"></img>
</td>
</tr>
</table>
<ul class="no-margin"><li>.darkblue[Benchmark Results]
  <ul><li>.darkblue[Reconfiguration] reduced from about half hour to few milliseconds (.darkblue[< 20 ms for all queries]) when using semi-static operator graphs</li>
  </ul></li>

<li>SP<sup>2</sup>B Benchmark
     <ul><li>Dataset sizes from 66 to 262 million triples</li>
     <li><span class="darkblue bold">Speedups between 4 and 32 times</span> <span class="darkgray">(dependent on query and dataset size)</span></li>
	 </ul>
  </li>
</ul></li>
</ul>
</div>

---
# Timeline of .darkblue[Quantum Computers]

.reference[<a href="https://en.wikipedia.org/wiki/List_of_quantum_processors" target="_blank">Main Data Source</a> Roadmaps: <a href="https://research.ibm.com/blog/ibm-quantum-roadmap" target="_blank">IBM</a> <a href="https://www.computerwoche.de/a/2025-kommt-der-erste-4-000-qubit-quantencomputer" target="_blank">IBM'25</a> <a href="https://quantumai.google/research" target="_blank">Google</a> <a href="https://doi.org/10.1007/s42354-021-0342-8" target="_blank">Quantum Brilliance</a> <a href="https://www.dwavesys.com/media/xvjpraig/clarity-roadmap_digital_v2.pdf" target="_blank">D-Wave</a> <a href="https://education.jlab.org/qa/mathatom_05.html" target="_blank">#Atoms on Earth</a> <a href="https://www.popularmechanics.com/space/a27259/how-many-particles-are-in-the-entire-universe/" target="_blank">#Particles in Universe</a>]

---
# Using .darkblue[Hardware Accelerator] for optimizing Queries / Transaction Schedules]

.reference[<a href="https://doi.org/10.1145/3410566.3410593" target="_blank">[BG20a]</a> <a href="http://nbn-resolving.de/urn:nbn:de:101:1-2020112218332015343957" target="_blank">[BG20b]</a>  <a href="https://doi.org/10.1145/3472163.3472164" target="_blank">[GG21]</a>]

---
# Approaches for .darkblue[Query/Transaction Schedule Optimization]

---
# .darkblue[Transaction Scheduling]

.reference[<a href="https://doi.org/10.14778/3598581.3598603" target="_blank">VLDB'23</a>]

.more-more-condensed.no-margin[
- Design choice: .darkblue[Fast quantum] exact methods .darkblue[and] good .darkblue[heuristics] for scalable optimization times
]
<table class="no-margin">
<tbody>
<tr>
<td>
.more-more-condensed.no-margin[
- .darkblue[Classical]
  - .darkblue[Dynamic Programming | Random Search | Simulated Annealing .darkgray[(SA)] | Genetic Algorithm .darkgray[(GA)] | Ant Colony Optimization .darkgray[(ACO)]]
- .darkblue[Quantum]
  - .darkblue[Maximum Independent Set | Quantum Evolutionary Algorithms] .darkgray[(QEA) (several variants: Binary | (Cultural (QCEA)) Multi-State)] .darkblue[| QUBO | Grover]
]
</td>
<td style="width:55%;vertical-align:top">
<div class="textcenter">
<img src="../../img/qc4db/TS_QEA.png" align="top" class="fullwidth"></img>
</div>
</td>
</tr>
</tbody>
</table>

---
# .darkblue[Transaction Scheduling] via .darkblue[Quantum Annealing]

.reference[<a href="https://doi.org/10.1145/3410566.3410593" target="_blank">T. Bittner, S. Groppe, Avoiding Blocking by Scheduling Transactions using Quantum Annealing, IDEAS 2020</a>]

.more-more-condensed.no-margin[
<table>
<tr>
<td style="width:65%;vertical-align:top">
.more-more-condensed.no-margin[
- Design Choice: Using .darkblue[Quantum Annealing for constant execution time]
- Experiments on real .darkblue[Quantum Annealer] (D-Wave 2000Q cloud service)
  - first minute free<br/>(afterwards too much for our budget)
- Versus .darkblue[Simulated Annealing on CPU]
- .darkblue[Preprocessing time/Number of QuBits]: $O((n\cdot k\cdot R)^2)$
]
</td><td style="width:35%;vertical-align:top">
<div class="textcenter"><img src="../../img/keynotes/ExperimentsQuantumAnnealing.svg" align="top" width="100%" class="padding-bottom"></img></div>
</td></tr>
</table>
<div class="textcenter no-margin"><img src="../../img/keynotes/ExperimentsQuantumAnnealingTable.svg" align="top" width="100%"></img><div class="smaller-font">$k$: #cores, $n$: #transactions, $R$: max. makespan, $O$: conflict matrix, $l_x$: transaction length, $r_x = R-l_x$</div></div>  
]

---
# .darkblue[QML] for .darkblue[Join Ordering]

.reference[Single-Step: <a href="https://doi.org/10.1145/3579142.3594299" target="_blank">BiDEDE@SIGMOD'23</a> <a href="https://doi.org/10.14778/3598581.3598603" target="_blank">VLDB'23</a> <a href="https://doi.org/10.1145/3555041.3589404" target="_blank">SIGMOD'23</a> Multi-Step (cooperation with OTH Regensburg): <a href="https://doi.org/10.48550/arXiv.2405.07770" target="_blank">QCE'24</a>]

.more-more-condensed.no-margin[
- Design choice: .darkblue[Predict best join order based on] real .darkblue[execution times] .darkgray[in contrast to estimated costs (EC)]
  - can .darkblue[beat exact methods] like dynamic programming (DP) based on EC
- Single-Step: .darkblue[Real-world queries] .darkgray[(ErgastF1 Benchmark/PostgreSQL)]
  - .darkblue[join orders with faster execution than DP (2.8%) and GEQO (11.2%)]
  - close to classical ML like RTOS (6%) / best join orders (16.8%)
]
<table>
<tbody>
<tr>
<td style="width:50%">
<div class="textcenter">
<img src="../../img/keynotes/bv/SingleVsMultiStepJoinOrdering.svg" align="top" style="width:95%"></img>
</div>
</td>
<td style="width:50%">
<div class="textcenter">
<img src="../../img/keynotes/bv/QMLJOO.svg" align="top" style="width:95%"></img>
</div>
</td>
</tr>
</tbody>
</table>

---
# .darkblue[Join Ordering as QUBO Problem]

.reference[<a href="https://doi.org/10.1145/3579142.3594298" target="_blank">BiDEDE@SIGMOD'23</a> Partitioning the Search Space: <a href="https://doi.org/10.1007/s13222-024-00468-3" target="_blank">Datenbank-Spektrum 24(1): 21-32, 2024</a>]

.more-more-condensed.no-margin[
- Design choice: .darkblue[Direct Join Costs] .darkgray[in contrast to calculated costs]
  - .darkblue[Estimated costs for .bold[each] join]  .darkblue[$\Rightarrow$ better join order]
  - .darkblue[Runtime] complexity .darkblue[$O(2^m-m)$ with $m$ relations] to join .darkblue[optimal] for join ordering based on direct costs
     - <span class="darkblue">Classical</span> runtime: <span class="darkblue">$O(3^m-2^{m+1})$</span> using dynamic programming
- .darkblue[Real-world queries] .darkgray[(ErgastF1 Benchmark/PostgreSQL)]
]
<table>
<tbody>
<tr>
<td style="width:55%">
<div class="textcenter">
<img src="../../img/keynotes/bv/QAResults.svg" align="top" style="width:95%"></img>
</div>
</td>
<td style="width:45%">
<div class="textcenter">
<img src="../../img/keynotes/bv/QAResultsSelectivity.svg" align="top" style="width:95%"></img>
</div>
</td>
</tr>
</tbody>
</table>

---
# .darkblue[QML for Cardinality Estimation]

.reference[MSCN (multi-set convolutional network): <a href="http://cidrdb.org/cidr2019/papers/p101-kipf-cidr19.pdf" target="_blank">Kipf et al. Learned cardinalities: Estimating correlated joins with deep learning, CIDR 2019</a>]

.more-more-condensed.no-margin[
- Design choice: .darkblue[Predict correction factor] of a Cardinality Estimator
  - .darkblue[Cardinality Predictor:PostgreSQL|MSCN|Hybrid Quantum Classical Network]
  - .darkblue[Cardinality Correction: Hybrid Quantum Classical Network]
- Queries of JOB-light Benchmark
]

---
# .darkblue[Transaction Scheduling Optimisation Server]

<img src="../../img/qc4db/TSOS.svg" align="top" class="fullwidth"></img>
.more-more-condensed[
- .darkblue[Demonstrator: fully quantum .bold[versus] fully classical]
]

---
# Demo for .darkblue[Query/Transaction Schedule Opt./Exec. Time Prediction]

.reference[Link to demo: <a href="https://qc4db.quantum-brilliance.com/" target="_blank">https://qc4db.quantum-brilliance.com/</a>]

.more-more-condensed[
- .darkblue[is publicly available] at the .darkblue[Quantum Brilliance web-page]...<br/>
<video class="fullwidth" controls>
  <source src="../../img/qc4db/quantum_demo.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
]

---
# .darkblue[Summary]
.more-more-condensed[
- .darkblue[Teaching]
  - .darkblue[50 lectures]
  - Easy-to-start .darkblue[tutorials for students for] much .darkblue[practical expertise]
  - Many .darkblue[supervisions] of students at PhD, master and bachelor level
  - .darkblue[Involving students in research] contributions (so far 25% of publications)
- .darkblue[Research]
  - .darkblue[$\approx$ 2M   <span class="material-icons"></span>project grants]
  - .darkblue[$>$ 191 publications]
  - .darkblue[Well integrated] in the top research community as shown by numerous scientific services
  - Main topics
     - <span class="darkblue">Advanced LLM Applications</span> (KG Construction, Chatbots, biodiversity, ...)
     - <span class="darkblue">Software Vulnerability Prediction</span>
     - <span class="darkblue">Green Computing</span>
     - <span class="darkblue">(Quantum) Machine Learning</span> for Query Optimization
     - <span class="darkblue">Quantum Computing</span> for Data Management
     - <span class="darkblue">Semantic Web</span>
]