feat: updated content

components/research/table.vue

@@ -7,7 +7,7 @@ const props = defineProps<{
 }>()
 
 const filters = ref();
-const statuses = ref(['conference', 'workshop', 'article', 'unpublished']);
+const statuses = ref(['conference', 'workshop', 'article', 'unpublished', 'thesis']);
 const loading = ref(false);
 
 const initFilters = () => {
@@ -40,6 +40,24 @@ const getStatus = (status) => {
             return 'warning';
     }
 };
+const getType = (type) => {
+    switch (type) {
+        case 'journal':
+            return 'success';
+
+        case 'workshop':
+            return 'info';
+
+        case 'seminar':
+            return 'warning';
+
+        case 'thesis':
+            return 'primary';
+
+        case 'article':
+            return 'danger';
+    }
+};
 
 const emit = defineEmits(['click']);
 
@@ -86,6 +104,11 @@ function onRowClick(event: any) {
                     {{ data.venue }}
                 </template>
             </Column>
+            <Column header="Type" field="type" sortable style="min-width: 6rem">
+                <template #body="{ data }">
+                    <Tag :value="data.type" :severity="getType(data.type)" />
+                </template>
+            </Column>
             <Column header="Status" field="status" sortable style="min-width: 6rem">
                 <template #body="{ data }">
                     <Tag :value="data.status" :severity="getStatus(data.status)" />

content/research/publications.json

@@ -1,12 +1,18 @@
 [
     {
-        "title": "Co-Adaptation of Agent Morphology and Behaviour with Self-Imitating Reinforcement Learning",
+        "title": "Following Ancestral Footsteps: Co-Designing Morphology and Behaviour with Self-Imitation Learning",
         "authors": ["Sergio Hernández-Gutiérrez", "Ville Kyrki", "Kevin S. Luck"],
-        "venue": "ICML",
+        "venue": "CoRL",
         "year": "2024",
         "status": "submitted",
-        "abstract": "In this paper we consider the problem of co-adapting the body and behaviour of agents, a long-standing research problem in the community of evolutionary robotics. Previous work has largely focused on the development of methods exploiting massive parallelization of agent evaluations with large population sizes, a paradigm which is not applicable to the real world. More recent data-efficient approaches utilizing reinforcement learning can suffer from distributional shifts to transition dynamics as well as to states and action spaces when experiencing new body morphologies. In this work, we propose a new co-adaptation method combining reinforcement learning and State-Aligned Self-Imitation Learning. We show that the integration of a self-imitation signal improves data-efficiency, behavioural recovery for unseen designs and performance convergence.",
+        "type": "journal",
+        "abstract": "In this paper we consider the problem of co-adapting the body and behaviour of agents, a long-standing research problem in the community of evolutionary robotics. Previous work has largely focused on the development of methods exploiting massive parallelization of agent evaluations with large population sizes, a paradigm which is not applicable to the real world. More recent data-efficient approaches utilizing reinforcement learning can suffer from distributional shifts in transition dynamics as well as in state and action spaces when experiencing new body morphologies. In this work, we propose a new co-adaptation method combining reinforcement learning and State-Aligned Self-Imitation Learning. We show that the integration of a self-imitation signal improves the data-efficiency of the co-adaptation process as well as the behavioural recovery when adapting morphological parameters.",
         "links": [
+            {
+                "title": "PDF",
+                "url": "https://drive.google.com/file/d/1m__d6i2WA7pgBz6B8TFKWo99KqeRt8wa/view?usp=sharing",
+                "icon": "tabler:file-type-pdf"
+            },
             {
                 "title": "GitHub",
                 "url": "https://github.com/serhez/cosil",
@@ -18,9 +24,10 @@
     {
         "title": "A Comprehensive Overview of Goal-Conditioned Hierarchical Reinforcement Learning: Algorithms, Challenges, and Future Directions",
         "authors": ["Sergio Hernández-Gutiérrez", "Vivienne Wang"],
-        "venue": "Seminar work",
+        "venue": "Aalto University",
         "year": "2023",
         "status": "unpublished",
+        "type": "seminar",
         "abstract": "Hierarchical reinforcement learning (HRL) methods have recently enabled higher sample efficiency in high-dimensional and long reinforcement learning (RL) problems. Goal-conditioned HRL (GCHRL) approaches concretize these hierarchical ideas by providing reachable sub-goals and considering a chain of policies that model the actions required to reach them, which are either less abstract sub-goals or the agent's native actions. This paper analyses and compares the current state-of-the-art GCHRL methods. Additionally, it discusses the current and future key challenges of the area, including efficient state space exploration, meaningful sub-goal generation and representation, the non-stationarity of policies and the transfer of skills learnt for one problem to solve another. Finally, it contributes to the current discussion on future directions and key focus points within the field of GCHRL.",
         "links": [
             {
@@ -31,16 +38,34 @@
         ]
     },
 
+    {
+        "title": "Modal Logic Theorem Provers and Validity Rates",
+        "authors": ["Sergio Hernández-Gutiérrez", "Robin Hirsch"],
+        "venue": "University College London",
+        "year": "2019",
+        "status": "published",
+        "type": "thesis",
+        "abstract": "During my Bachelor's thesis at UCL, supervised by Prof. Robin Hirsch, I carried out a study on the validity rates of modal logic formulae as their complexity increases (i.e., more allowed connectives and larger formulae). For this purpose, I implemented a frame-based analytical tableau theorem prover for propositional modal logics K, KT, KB, K4, KD and linear modal logic. This implementation was compared to Molle, a state-of-the-art theorem prover for modal logics at the time; this analysis found inconsistencies in the results of both provers, concluding with evidence of Molle's incorrectness on complex formulae.",
+        "links": [
+            {
+                "title": "PDF",
+                "url": "https://drive.google.com/file/d/1EUqOJb5ZK0bOM0Ix1LSmzKXcPGj2d8Pa/view?usp=sharing",
+                "icon": "tabler:file-type-pdf"
+            }
+        ]
+    },
+
     {
         "title": "3D Reconstruction of Fire-Damaged Parchments",
         "authors": ["Sergio Hernández-Gutiérrez", "Wanyue Zhang", "Ionut Deocanu"],
-        "venue": "Microsoft Blog, UK Faculty Connection",
+        "venue": "Microsoft Faculty Connection",
         "year": "2018",
         "status": "published",
-        "abstract": "In this post in partnership with Microsoft, as a Microsoft Student Partner, I give an introduction to 3D reconstruction of physical objects. In particular, I explain the process of reconstructing fire-damaged parchments and, as part of my 2nd year project at UCL, building a product for archivists and other professionals who are in need of a parchment-reconstruction tool to read them.",
+        "type": "article",
+        "abstract": "In this article in partnership with Microsoft, as a Microsoft Student Partner, I give an introduction to 3D reconstruction of physical objects. In particular, I explain the process of reconstructing fire-damaged parchments and, as part of my 2nd year project at UCL, building a product for archivists and other professionals who are in need of a parchment-reconstruction tool to read them.",
         "links": [
             {
-                "title": "Microsoft Blog",
+                "title": "Microsoft Faculty Connection",
                 "url": "https://learn.microsoft.com/en-us/archive/blogs/uk_faculty_connection/3d-reconstruction-of-fire-damage-parchments",
                 "icon": "mdi-microsoft"
             }

pages/index.vue

@@ -5,7 +5,11 @@ import emojis from '@/content/emojis.json';
 useHead({
     title: 'Sergio Hernández Gutiérrez',
     titleTemplate: '%s',
-    meta: [{ name: 'description', content: 'Sergio Hernández Gutiérrez, Machine Learning scientist.' }],
+    meta: [
+        {
+            name: 'description', content: 'Sergio Hernández Gutiérrez, Machine Learning scientist.'
+        }
+    ],
 });
 
 definePageMeta({
@@ -33,10 +37,10 @@ definePageMeta({
                 </div>
                 <div class="max-w-2xl space-y-4">
                     <p>I am a ML scientist currently looking for
-                        PhD opportunities starting from autumn 2024. I have experience designing and employing
-                        <strong>deep
-                            learning</strong> models, <strong>reinforcement learning</strong> and <strong>imitation
-                            learning</strong> approaches to solve complex <strong>reasoning problems</strong>. Check out
+                        PhD opportunities. My current research interests revolve around employing deep models to solve
+                        complex <strong>reasoning or decision-making tasks</strong>, how <strong>knowledge and data representation</strong> affects such
+                        capabilities, as well as learning the reinforcement learning process (<strong>meta-learning</strong>) outside of
+                        classical algorithmic approaches. Check out
                         my
                         <fancy-link to="/research">research output
                         </fancy-link>
@@ -46,7 +50,7 @@ definePageMeta({
                     <p>
                         Originally from the Canary Islands, I am currently finishing my <strong>MSc in ML at Aalto
                             University</strong>. I hold a <strong>BSc in Computer Science</strong> from
-                        <strong>University College London (UCL)</strong>. I have worked
+                        University College London (<strong>UCL</strong>). I have worked
                         as an
                         <strong>engineer/scientist for +3 years</strong> in Spain, Scotland and Finland, in
                         companies such as <strong>Rockstar Games</strong>, <strong>Cabify</strong> and
@@ -55,7 +59,7 @@ definePageMeta({
                     </p>
                 </div>
                 <div class="flex gap-6">
-                    <a v-for="social in socials" :key="social.title" :href="social.url" :title="social.title"
+                    <a v-for="social in socials":key="social.title":href="social.url":title="social.title"
                         class="text-zinc-600 transition hover:text-zinc-300" rel="me nofollow noopener" target="_blank">
                         <icon :name="social.icon" class="h-6
 w-6" />

pages/research.vue

@@ -47,7 +47,7 @@ function togglePopup(publication: any) {
             </div>
             <div class="flex flex-col gap-0">
                 <h1>
-                    Publications
+                    Research output
                 </h1>
                 <research-table class="w-full" :publications="publications" @click="togglePopup" />
             </div>

pages/resume.vue

@@ -9,7 +9,8 @@
                         University,
                         Finland</h2>
                     <p>2022 — In progress</p>
-                    <p>Currently a <strong>GPA</strong> of 5.0 out of 5.0.</p>
+                    <p>Currently a <strong>GPA</strong> of 5.0 out of 5.0 (max. grade in all classes), pending my
+                        thesis.</p>
                 </div>
                 <div class="flex flex-col gap-1">
                     <h2><strong>BSc Computer Science</strong> — University College London, UK</h2>
@@ -31,13 +32,11 @@
                         <strong>reinforcement learning co-adaptation
                             setting</strong> (i.e., the design and behavior of an agent are jointly optimized). We have
                         observed that
-                        including this signal <strong>improves performance by 15-20%</strong> in terms of achieved
-                        rewards.
+                        including this signal <strong>improves achieved rewards</strong>; these benefits scale
+                        proportionally to the complexity of the task.
                         I gave a <strong>40-min.
-                            talk</strong> at the department about the project. This work culminated in a paper
-                        co-authored
-                        with Prof.
-                        Kyrki and Prof. Luck, which we have <strong>submitted for ICML 2024</strong>.
+                            talk</strong> at the department about the project. This work culminated in a paper, which we
+                        have <strong>submitted for CoRL 2024</strong>.
                     </p>
                 </div>
                 <div class="flex flex-col gap-1">
@@ -82,7 +81,7 @@
                 <h1>Courses (non-exhaustive)</h1>
                 <div class="flex flex-row gap-4">
                     <div class="flex flex-col gap-1 basis-1/2">
-                        <h2>Post-graduate</h2>
+                        <h2>Graduate</h2>
                         <ul class="list-disc pl-6">
                             <li>Statistical NLP</li>
                             <li>Reinforcement Learning</li>
@@ -99,7 +98,7 @@
                         </ul>
                     </div>
                     <div class="flex flex-col gap-1 basis-1/2">
-                        <h2>Graduate</h2>
+                        <h2>Undergraduate</h2>
                         <ul class="list-disc pl-6">
                             <li>Discrete Mathematics</li>
                             <li>Mathematics & Statistics</li>