CERCLE - Cercle d'Arquitecturahttp://hdl.handle.net/2117/31472024-03-28T11:41:59Z2024-03-28T11:41:59ZConcurs per a la redacció dels Projectes bàsic i executiu corresponent a la construcció d'una Escola bressol i l'ampliació de l'Escola Arnau Berenguer a El Palau d'AnglesolaJiménez Gómez, EvaLlobet Ribeiro, XavierZubelzu Viarje, Anahttp://hdl.handle.net/2117/3635722022-03-08T14:00:57Z2022-03-08T13:58:59ZConcurs per a la redacció dels Projectes bàsic i executiu corresponent a la construcció d'una Escola bressol i l'ampliació de l'Escola Arnau Berenguer a El Palau d'Anglesola
Jiménez Gómez, Eva; Llobet Ribeiro, Xavier; Zubelzu Viarje, Ana
Ajuntament del Palau d'Anglesola
2022-03-08T13:58:59ZJiménez Gómez, EvaLlobet Ribeiro, XavierZubelzu Viarje, AnaThe dark side of DNN pruningYazdani Aminabadi, RezaArnau Montañés, José MaríaGonzález Colás, Antonio MaríaRiera Villanueva, Marchttp://hdl.handle.net/2117/1251412021-09-05T12:27:49Z2018-11-27T18:23:47ZThe dark side of DNN pruning
Yazdani Aminabadi, Reza; Arnau Montañés, José María; González Colás, Antonio María; Riera Villanueva, Marc
DNN pruning has been recently proposed as an effective technique to improve the energy-efficiency of DNN-based solutions. It is claimed that by removing unimportant or redundant connections, the pruned DNN delivers higher performance and energy-efficiency with negligible impact on accuracy. However, DNN pruning has an important side effect: it May reduce the confidence of DNN predictions. We show that, although top-1 accuracy May be maintained with DNN pruning, the likelihood of the class in the top-1 is significantly reduced when using the pruned models. For applications such as Automatic Speech Recognition (ASR), where the DNN scores are consumed by a successive stage, the workload of this stage can be dramatically increased due to the loss of confidence in the DNN. An ASR system consists of a DNN for computing acoustic scores, followed by a Viterbi beam search to find the most likely sequence of words. We show that, when pruning the DNN model used for acoustic scoring, the Word Error Rate (WER) is maintained but the execution time of the ASR system is increased by 33%. Although pruning improves the efficiency of the DNN, it results in a huge increase of activity in the Viterbi search since the output scores of the pruned model are less reliable. Based on this observation, we propose a novel hardware-based ASR system that effectively integrates a DNN accelerator for pruned models with a Viterbi accelerator. In order to avoid the aforementioned increase in Viterbi search workload, our system loosely selects the N-best hypotheses at every time step, exploring only the N most likely paths. To avoid an expensive sort of the hypotheses based on their likelihoods, our accelerator employs a set-associative hash table to keep track of the best paths mapped to each set. In practice, this solution approaches the selection of N-best, but it requires much simpler hardware. Our approach manages to efficiently combine both DNN pruning and Viterbi search, and achieves 9x energy savings and 4.2x speedup with respect to the state-of-the-art ASR solutions.
2018-11-27T18:23:47ZYazdani Aminabadi, RezaArnau Montañés, José MaríaGonzález Colás, Antonio MaríaRiera Villanueva, MarcDNN pruning has been recently proposed as an effective technique to improve the energy-efficiency of DNN-based solutions. It is claimed that by removing unimportant or redundant connections, the pruned DNN delivers higher performance and energy-efficiency with negligible impact on accuracy. However, DNN pruning has an important side effect: it May reduce the confidence of DNN predictions. We show that, although top-1 accuracy May be maintained with DNN pruning, the likelihood of the class in the top-1 is significantly reduced when using the pruned models. For applications such as Automatic Speech Recognition (ASR), where the DNN scores are consumed by a successive stage, the workload of this stage can be dramatically increased due to the loss of confidence in the DNN. An ASR system consists of a DNN for computing acoustic scores, followed by a Viterbi beam search to find the most likely sequence of words. We show that, when pruning the DNN model used for acoustic scoring, the Word Error Rate (WER) is maintained but the execution time of the ASR system is increased by 33%. Although pruning improves the efficiency of the DNN, it results in a huge increase of activity in the Viterbi search since the output scores of the pruned model are less reliable. Based on this observation, we propose a novel hardware-based ASR system that effectively integrates a DNN accelerator for pruned models with a Viterbi accelerator. In order to avoid the aforementioned increase in Viterbi search workload, our system loosely selects the N-best hypotheses at every time step, exploring only the N most likely paths. To avoid an expensive sort of the hypotheses based on their likelihoods, our accelerator employs a set-associative hash table to keep track of the best paths mapped to each set. In practice, this solution approaches the selection of N-best, but it requires much simpler hardware. Our approach manages to efficiently combine both DNN pruning and Viterbi search, and achieves 9x energy savings and 4.2x speedup with respect to the state-of-the-art ASR solutions.Air movement in the ventilated façade: analysis of cases in Barcelona, SpainBalter, JulietaPardal March, CristinaCoch Roura, HelenaGanem, CarolinaParicio Ansuátegui, Ignaciohttp://hdl.handle.net/2117/1227682024-03-10T11:35:37Z2018-10-22T17:40:36ZAir movement in the ventilated façade: analysis of cases in Barcelona, Spain
Balter, Julieta; Pardal March, Cristina; Coch Roura, Helena; Ganem, Carolina; Paricio Ansuátegui, Ignacio
The Ventilated Façade (VF) supposes an improvement compared with a conventional façade –single layerin terms of thermal comfort, through the ventilation of the wall cavity, limited by an inner and an outer layer,
usually of open joints. However, in buildings with VF the constructions characteristics in the wall cavity are often different from that considered in academic studies, performed by simulations or prototype measurements. The present work analyses the air movement in the VF wall cavity of existing buildings built in Barcelona city. In situ measurements were made in summer in 10 buildings, using a Testo405i thermal anemometer with management via a smartphone. The methodology consisted in the monitoring of the cases in periods of 30 minutes in the morning and 30 minutes in the afternoon. The data were recorded every two seconds, according to the sequence: 5 minutes outside, 20 minutes inside the wall cavity and 5 minutes outside. The extensible telescope (400mm long) of the anemometer was introduced by the open joints of the outer sheets, so that the sensor is perpendicular to the vertical air flow in the wall cavity. The results show
two determining factors for the adequate performance of the air movement in the VF chamber: on the one hand, the increase in the incidence of solar radiation on the facade, and on the other hand, the increase of the entrance and exit apertures of air in the chamber. Besides, temperatures tend to increase in the wall cavity, that is to say that there are always gains of heat in the air chamber.
2018-10-22T17:40:36ZBalter, JulietaPardal March, CristinaCoch Roura, HelenaGanem, CarolinaParicio Ansuátegui, IgnacioThe Ventilated Façade (VF) supposes an improvement compared with a conventional façade –single layerin terms of thermal comfort, through the ventilation of the wall cavity, limited by an inner and an outer layer,
usually of open joints. However, in buildings with VF the constructions characteristics in the wall cavity are often different from that considered in academic studies, performed by simulations or prototype measurements. The present work analyses the air movement in the VF wall cavity of existing buildings built in Barcelona city. In situ measurements were made in summer in 10 buildings, using a Testo405i thermal anemometer with management via a smartphone. The methodology consisted in the monitoring of the cases in periods of 30 minutes in the morning and 30 minutes in the afternoon. The data were recorded every two seconds, according to the sequence: 5 minutes outside, 20 minutes inside the wall cavity and 5 minutes outside. The extensible telescope (400mm long) of the anemometer was introduced by the open joints of the outer sheets, so that the sensor is perpendicular to the vertical air flow in the wall cavity. The results show
two determining factors for the adequate performance of the air movement in the VF chamber: on the one hand, the increase in the incidence of solar radiation on the facade, and on the other hand, the increase of the entrance and exit apertures of air in the chamber. Besides, temperatures tend to increase in the wall cavity, that is to say that there are always gains of heat in the air chamber.A novel register renaming technique for out-of-order processorsTabani, HamidArnau Montañés, José MaríaTubella Murgadas, JordiGonzález Colás, Antonio Maríahttp://hdl.handle.net/2117/1222642022-05-17T11:17:18Z2018-10-11T14:41:14ZA novel register renaming technique for out-of-order processors
Tabani, Hamid; Arnau Montañés, José María; Tubella Murgadas, Jordi; González Colás, Antonio María
Modern superscalar processors support a large number of in-flight instructions, which requires sizeable register files. Conventional register renaming techniques allocate a new storage location, i.e. physical register, for every instruction whose destination is a logical register in order to remove false dependences. Physical registers are released in a conservative manner when the same logical register is redefined. For this reason, many cycles may happen between the last read and the release of a physical register, leading to suboptimal utilization of the register file. We have observed that for more than 50% of the instructions in SPECfp and more than 30% of the instructions in SPECint that have a destination register, the produced value has only a single consumer. In this case, the RAW dependence guarantees that the producer-consumer instructions pair will be executed in program order and, hence, the same physical register can be used to store the value produced by both instructions. In this paper, we propose a renaming technique that exploits this property to reduce the pressure on the register file. Our technique leverages physical register sharing by introducing minor changes in the register map table and the issue queue. We also describe how our renaming scheme supports precise exceptions. We evaluated our renaming technique on top of a modern out-of-order processor. Our experimental results show that it provides 6% speedup on average for the SPEC2006 benchmarks. Alternatively, our renaming scheme achieves the same performance while reducing the number of physical registers by 10.5%.
2018-10-11T14:41:14ZTabani, HamidArnau Montañés, José MaríaTubella Murgadas, JordiGonzález Colás, Antonio MaríaModern superscalar processors support a large number of in-flight instructions, which requires sizeable register files. Conventional register renaming techniques allocate a new storage location, i.e. physical register, for every instruction whose destination is a logical register in order to remove false dependences. Physical registers are released in a conservative manner when the same logical register is redefined. For this reason, many cycles may happen between the last read and the release of a physical register, leading to suboptimal utilization of the register file. We have observed that for more than 50% of the instructions in SPECfp and more than 30% of the instructions in SPECint that have a destination register, the produced value has only a single consumer. In this case, the RAW dependence guarantees that the producer-consumer instructions pair will be executed in program order and, hence, the same physical register can be used to store the value produced by both instructions. In this paper, we propose a renaming technique that exploits this property to reduce the pressure on the register file. Our technique leverages physical register sharing by introducing minor changes in the register map table and the issue queue. We also describe how our renaming scheme supports precise exceptions. We evaluated our renaming technique on top of a modern out-of-order processor. Our experimental results show that it provides 6% speedup on average for the SPEC2006 benchmarks. Alternatively, our renaming scheme achieves the same performance while reducing the number of physical registers by 10.5%.Esa escuela del VallèsParicio Ansuátegui, IgnacioPardal March, Cristinahttp://hdl.handle.net/2117/1137572024-03-10T13:52:11Z2018-02-05T14:36:22ZEsa escuela del Vallès
Paricio Ansuátegui, Ignacio; Pardal March, Cristina
A través de un minucioso proceso de investigación y diseño, cada proyecto de Harquitectes busca la coherencia entre imagen y funcionamiento energético.
Through a detailed process of research and design, each project by Harquitectes seeks to achieve coherence between image and energy performance.
2018-02-05T14:36:22ZParicio Ansuátegui, IgnacioPardal March, CristinaA través de un minucioso proceso de investigación y diseño, cada proyecto de Harquitectes busca la coherencia entre imagen y funcionamiento energético.
Through a detailed process of research and design, each project by Harquitectes seeks to achieve coherence between image and energy performance.Redacció del Pla Especial de Millora Urbana de l’Illa Glòries i de quatre promocions d’habitatge públic a l’Illa GlòriesValor Montero, JaumeLlimós Aicart, Laurahttp://hdl.handle.net/2117/1130782020-07-22T17:45:48Z2018-01-22T19:11:41ZRedacció del Pla Especial de Millora Urbana de l’Illa Glòries i de quatre promocions d’habitatge públic a l’Illa Glòries
Valor Montero, Jaume; Llimós Aicart, Laura
Ajuntament de Barcelona
2018-01-22T19:11:41ZValor Montero, JaumeLlimós Aicart, LauraBiblioteca de Sant Martí SarrocaValor Montero, JaumeLlimós Aicart, Laurahttp://hdl.handle.net/2117/1130772020-07-22T17:46:27Z2018-01-22T19:01:29ZBiblioteca de Sant Martí Sarroca
Valor Montero, Jaume; Llimós Aicart, Laura
Diputació de Barcelona
2018-01-22T19:01:29ZValor Montero, JaumeLlimós Aicart, LauraInstitutValor Montero, JaumeLlimós Aicart, Laurahttp://hdl.handle.net/2117/1130752020-07-22T17:50:36Z2018-01-22T18:52:10ZInstitut
Valor Montero, Jaume; Llimós Aicart, Laura
Departament d'Ensenyament de la Generalitat de Catalunya - GISA
2018-01-22T18:52:10ZValor Montero, JaumeLlimós Aicart, LauraAmpliació de l’Institut “Badia i Margarit” a IgualadaValor Montero, JaumeLlimós Aicart, Laurahttp://hdl.handle.net/2117/1130732020-07-22T17:42:13Z2018-01-22T18:44:30ZAmpliació de l’Institut “Badia i Margarit” a Igualada
Valor Montero, Jaume; Llimós Aicart, Laura
Departament d'Ensenyament de la Generalitat de Catalunya - GISA
2018-01-22T18:44:30ZValor Montero, JaumeLlimós Aicart, LauraAmpliació de l’Institut de Mollet del VallèsValor Montero, JaumeLlimós Aicart, Laurahttp://hdl.handle.net/2117/1130702020-07-22T17:34:13Z2018-01-22T18:31:15ZAmpliació de l’Institut de Mollet del Vallès
Valor Montero, Jaume; Llimós Aicart, Laura
Departament d'Ensenyament de la Generalitat de Catalunya - GISA
2018-01-22T18:31:15ZValor Montero, JaumeLlimós Aicart, Laura