Publications with Fulltext
Permanent URI for this collectionhttps://hdl.handle.net/20.500.14288/6
Browse
64 results
Search Results
Publication Open Access Engagement rewarded actor-critic with conservative Q-learning for speech-driven laughter backchannel generation(Association for Computing Machinery (ACM), 2021) Department of Computer Engineering; Bayramoğlu, Öykü Zeynep; Erzin, Engin; Sezgin, Tevfik Metin; Yemez, Yücel; Faculty Member; Faculty Member; Faculty Member; Department of Computer Engineering; Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); College of Engineering; Graduate School of Sciences and Engineering; N/A; 34503; 18632; 107907We propose a speech-driven laughter backchannel generation model to reward engagement during human-agent interaction. We formulate the problem as a Markov decision process where speech signal represents the state and the objective is to maximize human engagement. Since online training is often impractical in the case of human-agent interaction, we utilize the existing human-to-human dyadic interaction datasets to train our agent for the backchannel generation task. We address the problem using an actor-critic method based on conservative Q-learning (CQL), that mitigates the distributional shift problem by suppressing Q-value over-estimation during training. The proposed CQL based approach is evaluated objectively on the IEMOCAP dataset for laughter generation task. When compared to the existing off-policy Q-learning methods, we observe an improved compliance with the dataset in terms of laugh generation rate. Furthermore, we show the effectiveness of the learned policy by estimating the expected engagement using off-policy policy evaluation techniques.Publication Open Access The structural basis of Akt PH domain interaction with calmodulin(Elsevier, 2021) Jang, Hyunbum; Nussinov, Ruth; N/A; Department of Chemical and Biological Engineering; Department of Computer Engineering; Weako, Jackson; Keskin, Özlem; Gürsoy, Attila; Faculty Member; Department of Chemical and Biological Engineering; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; N/A; 26605; 8745Akt plays a key role in the Ras/PI3K/Akt/mTOR signaling pathway. In breast cancer, Akt translocation to the plasma membrane is enabled by the interaction of its pleckstrin homology domain (PHD) with calmodulin (CaM). At the membrane, the conformational change promoted by PIP3 releases CaM and facilitates Thr308 and Ser473 phosphorylation and activation. Here, using modeling and molecular dynamics simulations, we aim to figure out how CaM interacts with Akt's PHD at the atomic level. Our simulations show that CaM-PHD interaction is thermodynamically stable and involves a beta-strand rather than an alpha-helix, in agreement with NMR data, and that electrostatic and hydrophobic interactions are critical. The PHD interacts with CaM lobes; however, multiple modes are possible. IP4, the polar head of PIP3, weakens the CaM-PHD interaction, implicating the release mechanism at the plasma membrane. Recently, we unraveled the mechanism of PI3K alpha activation at the atomistic level and the structural basis for Ras role in the activation. Here, our atomistic structural data clarify the mechanism of how CaM interacts, delivers, and releases Akt-the next node in the Ras/PI3K pathway-at the plasma membrane.Publication Open Access Craft: a benchmark for causal reasoning about forces and in teractions(Association for Computational Linguistics (ACL), 2022) Ateş, Tayfun; Ateşoğlu, M. Şamil; Yiğit, Çağatay; Department of Computer Engineering; Department of Psychology; Erdem, Aykut; Göksun, Tilbe; Yüret, Deniz; Kesen, İlker; Kobaş, Mert; Faculty Member; Faculty Member; Faculty Member; Master Student; Department of Computer Engineering; Department of Psychology; Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); Graduate School of Sciences and Engineering; College of Engineering; College of Social Sciences and Humanities; 20331; 47278; 179996; N/A; N/A; N/AHumans are able to perceive, understand and reason about causal events. Developing models with similar physical and causal understanding capabilities is a long-standing goal of artificial intelligence. As a step towards this direction, we introduce CRAFT1, a new video question answering dataset that requires causal reasoning about physical forces and object interactions. It contains 58K video and question pairs that are generated from 10K videos from 20 different virtual environments, containing various objects in motion that interact with each other and the scene. Two question categories in CRAFT include previously studied descriptive and counterfactual questions. Additionally, inspired by the Force Dynamics Theory in cognitive linguistics, we introduce a new causal question category that involves understanding the causal interactions between objects through notions like cause, enable, and prevent. Our results show that even though the questions in CRAFT are easy for humans, the tested baseline models, including existing state-of-the-art methods, do not yet deal with the challenges posed in our benchmark.Publication Open Access Kart-ON: an extensible paper programming strategy for affordable early programming education(Association for Computing Machinery (ACM), 2022) Department of Computer Engineering; Sezgin, Tevfik Metin; Sabuncuoğlu, Alpay; Faculty Member; Department of Computer Engineering; Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); College of Engineering; Graduate School of Sciences and Engineering; 18632; N/AProgramming has become a core subject in primary and middle school curricula. Yet, conventional solutions for in-class programming activities require each student to have expensive equipment, which creates an opportunity gap for low-income students. Paper programming can provide an affordable, engaging, and collaborative in-class programming experience by allowing groups of students to use inexpensive materials and share smartphones. However, current paper-programming examples are limited in terms of language expressivity and generalizability. Addressing these limitations, we developed a paper-programming flow and its variants in different abstraction levels and input/output styles. The programming environments consist of pre-defined tangible programming cards and a mobile application that runs computer vision models to recognize them. This paper describes our educational and technical development process, presents a qualitative analysis of the early user study results and shares our design considerations to help develop wide-reaching paper programming environments.Publication Open Access Interlaced: fully decentralized churn stabilization for Skip Graph-based DHTs(Elsevier, 2021) Department of Computer Engineering; Hassanzadeh-Nazarabadi, Yahya; Küpçü, Alptekin; Özkasap, Öznur; PhD Student; Faculty Member; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; N/A; 168060; 113507As a distributed hash table (DHT) routing overlay, Skip Graph is used in a variety of peer-to-peer (P2P) systems including cloud storage. The overlay connectivity of P2P systems is negatively affected by the arrivals and departures of nodes to and from the system that is known as churn. Preserving connectivity of the overlay network (i.e., the reachability of every pair of nodes) under churn without compromising the overlay latency is a performance challenge in every P2P system including the Skip Graph-based ones. The existing decentralized churn stabilization solutions that are applicable to Skip Graphs mainly optimize the connectivity of the system under churn and do not consider routing latency of overlay as an optimization goal. Additionally, those existing solutions change the message complexity of Skip Graphs, distort its topology, or apply constant message overhead to the system. In this paper, we propose Interlaced, a fully decentralized churn stabilization mechanism for Skip Graphs that provides drastically stronger overlay connectivity and faster search queries without changing the asymptotic complexity of the Skip Graph in terms of storage, computation, and communication. We also propose the Sliding Window De Bruijn Graph (SWDBG ) as a tool to predict the availability of nodes with high accuracy. Our simulation results show that in comparison to the best existing DHT-based solutions, Interlaced improves the overlay connectivity of the Skip Graph under churn with the gain of about 1.73 times. Likewise, compared to the existing availability prediction approaches for P2P systems, SWDBG is about 1.26 times more accurate. A Skip Graph that benefits from Interlaced and SWDBG is about 2.47 times faster on average in routing the queries under churn compared to the best existing solutions. We also present an adaptive extension of Interlaced to be applied to other DHTs, for example, Kademlia.Publication Open Access Leveraging frequency based salient spatial sound localization to improve 360 degrees video saliency prediction(Institute of Electrical and Electronics Engineers (IEEE), 2021) Çökelek, Mert; İmamoğlu, Nevrez; Özçınar, Çağrı; Department of Computer Engineering; Erdem, Aykut; Faculty Member; Department of Computer Engineering; College of Engineering; 20331Virtual and augmented reality (VR/AR) systems dramatically gained in popularity with various application areas such as gaming, social media, and communication. It is therefore a crucial task to have the knowhow to efficiently utilize, store or deliver 360° videos for end-users. Towards this aim, researchers have been developing deep neural network models for 360° multimedia processing and computer vision fields. In this line of work, an important research direction is to build models that can learn and predict the observers' attention on 360° videos to obtain so-called saliency maps computationally. Although there are a few saliency models proposed for this purpose, these models generally consider only visual cues in video frames by neglecting audio cues from sound sources. In this study, an unsupervised frequency-based saliency model is presented for predicting the strength and location of saliency in spatial audio. The prediction of salient audio cues is then used as audio bias on the video saliency predictions of state-of-the-art models. Our experiments yield promising results and show that integrating the proposed spatial audio bias into the existing video saliency models consistently improves their performance.Publication Open Access Mechanistic differences of activation of Rac1(P29S) and Rac1(A159V)(American Chemical Society (ACS), 2021) Jang, Hyunbum; Nussinov, Ruth; N/A; Department of Chemical and Biological Engineering; Department of Computer Engineering; Şenyüz, Simge; Keskin, Özlem; Gürsoy, Attila; Faculty Member; Department of Chemical and Biological Engineering; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; N/A; 26605; 8745Rac1 is a small GTPase that plays key roles in actin reorganization, cell motility, and cell survival/growth as well as in various cancer types and neurodegenerative diseases. Similar to other Ras superfamily GTPases, Rac1 switches between active GTP-bound and inactive GDP-bound states. Switch I and II regions open and close during GDP/GTP exchange. P29S and A159V (paralogous to K-Ras(A146)) mutations are the two most common somatic mutations of Rac1. Rac1(P2)(9S)( )is a known hotspot for melanoma, whereas Rac1(A159V) most commonly occurs in head and neck cancer. To investigate how these substitutions induce the Rac1 dynamics, we used atomistic molecular dynamics simulations on the wild-type Rac1 and two mutant systems (P29S and A159V) in the GTP bound state, and on the wild-type Rac1 and P29S mutated system in the GDP bound state. Here, we show that P29S and A159V mutations activate Rac1 with different mechanisms. In Rac1(P29S)-GTP, the substitution increases the flexibility of Switch I based on RMSF and dihedral angle calculations and leads to an open conformation. We propose that the open Switch I conformation is one of the underlying reasons for rapid GDP/GTP exchange of Rac1(P29S). On the other hand, in Rac1(A159V)-GTP, some of the contacts of the guanosine ring of GTP with Rac1 are temporarily lost, enabling the guanosine ring to move toward Switch I and subsequently close the switch. Rac1(A159V)-GTP adopts a Ras state 2 like conformation, where both switch regions are in closed conformation and Thr35 forms a hydrogen bond with the nucleotide.Publication Open Access Androgen receptor-mediated transcription in prostate cancer(Multidisciplinary Digital Publishing Institute (MDPI), 2022) Morova, Tunç; Department of Computer Engineering; Department of Chemical and Biological Engineering; Lack, Nathan Alan; Özturan, Doğancan; Faculty Member; PhD Student; Department of Computer Engineering; Department of Chemical and Biological Engineering; Koç University Research Center for Translational Medicine (KUTTAM) / Koç Üniversitesi Translasyonel Tıp Araştırma Merkezi (KUTTAM); School of Medicine; 120842; N/AAndrogen receptor (AR)-mediated transcription is critical in almost all stages of prostate cancer (PCa) growth and differentiation. This process involves a complex interplay of coregulatory proteins, chromatin remodeling complexes, and other transcription factors that work with AR at cis-regulatory enhancer regions to induce the spatiotemporal transcription of target genes. This enhancer-driven mechanism is remarkably dynamic and undergoes significant alterations during PCa progression. In this review, we discuss the AR mechanism of action in PCa with a focus on how cis-regulatory elements modulate gene expression. We explore emerging evidence of genetic variants that can impact AR regulatory regions and alter gene transcription in PCa. Finally, we highlight several outstanding questions and discuss potential mechanisms of this critical transcription factor.Publication Open Access Self-supervised monocular scene decomposition and depth estimation(IEEE Computer Society, 2021) Department of Computer Engineering; N/A; Güney, Fatma; Safadoust, Sadra; Department of Computer Engineering; Koç Üniversitesi İş Bankası Yapay Zeka Uygulama ve Araştırma Merkezi (KUIS AI)/ Koç University İş Bank Artificial Intelligence Center (KUIS AI); College of Engineering; Graduate School of Sciences and Engineering; 187939; N/ASelf-supervised monocular depth estimation approaches either ignore independently moving objects in the scene or need a separate segmentation step to identify them. We propose MonoDepthSeg to jointly estimate depth and segment moving objects from monocular video without using any ground-truth labels. We decompose the scene into a fixed number of components where each component corresponds to a region on the image with its own transformation matrix representing its motion. We estimate both the mask and the motion of each component efficiently with a shared encoder. We evaluate our method on three driving datasets and show that our model clearly improves depth estimation while decomposing the scene into separately moving components.Publication Open Access A diversity combination model incorporating an inward bias for interaural time-level difference cue integration in sound lateralization(Multidisciplinary Digital Publishing Institute (MDPI), 2020) N/A; Department of Computer Engineering; Mojtahedi, Sina; Erzin, Engin; Ungan, Pekcan; Faculty Member; Faculty Member; Department of Computer Engineering; Graduate School of Sciences and Engineering; College of Engineering; School of Medicine; N/A; 34503; N/AA sound source with non-zero azimuth leads to interaural time level differences (ITD and ILD). Studies on hearing system imply that these cues are encoded in different parts of the brain, but combined to produce a single lateralization percept as evidenced by experiments indicating trading between them. According to the duplex theory of sound lateralization, ITD and ILD play a more significant role in low-frequency and high-frequency stimulations, respectively. In this study, ITD and ILD, which were extracted from a generic head-related transfer functions, were imposed on a complex sound consisting of two low- and seven high-frequency tones. Two-alternative forced-choice behavioral tests were employed to assess the accuracy in identifying a change in lateralization. Based on a diversity combination model and using the error rate data obtained from the tests, the weights of the ITD and ILD cues in their integration were determined by incorporating a bias observed for inward shifts. The weights of the two cues were found to change with the azimuth of the sound source. While the ILD appears to be the optimal cue for the azimuths near the midline, the ITD and ILD weights turn to be balanced for the azimuths far from the midline.