Sense and Learn: Self-supervision for omnipresent sensors
نویسندگان
چکیده
Learning general-purpose representations from multisensor data produced by the omnipresent sensing systems (or IoT in general) has numerous applications diverse use cases. Existing purely supervised end-to-end deep learning techniques depend on availability of a massive amount well-curated data, acquiring which is notoriously difficult but required to achieve sufficient level generalization task interest. In this work, we leverage self-supervised paradigm towards realizing vision continual unlabeled inputs. We present generalized framework named Sense and Learn for representation or feature raw sensory data. It consists several auxiliary tasks that can learn high-level broadly useful features entirely unannotated without any human involvement tedious labeling process. demonstrate efficacy our approach publicly available datasets different domains various settings, including linear separability, semi-supervised few shot learning, transfer learning. Our methodology achieves results are competitive with approaches close gap through fine-tuning network while downstream most particular, show be utilized as initialization significantly boost performance low-data regime 5 labeled instances per class, high practical importance real-world problems. Likewise, learned self-supervision found highly transferable between related datasets, even when target domains.
منابع مشابه
the search for the self in becketts theatre: waiting for godot and endgame
this thesis is based upon the works of samuel beckett. one of the greatest writers of contemporary literature. here, i have tried to focus on one of the main themes in becketts works: the search for the real "me" or the real self, which is not only a problem to be solved for beckett man but also for each of us. i have tried to show becketts techniques in approaching this unattainable goal, base...
15 صفحه اولLearning to Learn from Weak Supervision by Full Supervision
In this paper, we propose a method for training neural networks when we have a large set of data with weak labels and a small amount of data with true labels. In our proposed model, we train two neural networks: a target network, the learner and a confidence network, the meta-learner. The target network is optimized to perform a given task and is trained using a large set of unlabeled data that...
متن کاملSelf-Supervision for Reinforcement Learning
Reinforcement learning optimizes policies for expected cumulative reward. Need the supervision be so narrow? Reward is delayed and sparse for many tasks, making it a difficult and impoverished signal for end-to-end optimization. To augment reward, we consider a range of self-supervised tasks that incorporate states, actions, and successors to provide auxiliary losses. These losses offer ubiquit...
متن کاملan investigation into elt teachers’ sense of self-efficacy in teaching reading comprehension and their personality types
as far as bandura’s (1977) conceptualization of the sense of self-efficacy is concerned, it is argued to be context- and even task-specific, compared to other concepts like self-esteem (brown, 2000); therefore, upon the previous research studies, related theories and definitions, and the ideas and views obtained from the interviews conducted with the scholars, the current research study aimed a...
ذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Machine learning with applications
سال: 2021
ISSN: ['2666-8270']
DOI: https://doi.org/10.1016/j.mlwa.2021.100152