No description available.

Master of Agribusiness / Department of Agricultural Economics / Major Professor Not Listed / This thesis was written as a business plan for the purpose of securing funds for a new micro distillery. Wheat State Distilling is an artisan distillery, manufacturing and selling premium hand-crafted spirits from the regions finest ingredients. I plan to distill premium vodka, whisky, bourbon, rum, gin and liqueurs as well as a luxury line of reserve products. Kansas established a micro-distillery license in July 2012 that opens up the market to this new type of business. The micro-distillery trend is growing across the U.S. and I plan to establish myself as a key player in the region. I will be entering the market early in the industry’s life cycle when consumer interest and demand are high. The proposed location in Wichita Kansas takes advantage of the largest city in Kansas with a metro area population of 630,721 in 2011. The location has visible signage from Interstate 135. The distillery will offer tours and a tasting room. We are able to give customers free samples and sell cocktails to induce purchase. We will use direct sales as well as restaurant and liquor store sales thru distributors. I will be marketing my premium products to upper-middle class and wealthy consumers. My extensive product line will allow marketing to a wide age range and to both sexes. We will focus on online and social media as well as distributor incentives and in-store promotions. Participation in local and regional events for sampling and advertising will increases brand awareness. Analysis of monthly production levels of each of the proposed product has been completed to maximize profits considering time and equipment constraints, cost of inputs and sale price. Regression analysis has been completed examining many factors that contribute to the selling price of micro distilled spirits. Financial models for years one through ten with risk analysis are included. If I reach my sales goals of 25,000 cases per year by year five, which is equal to my closest competition, High Plains Distilling, there is a potential for $548,500 or more per year in profit. I am seeking a loan for $300,000.

No description available.

Typescript (photocopy). / Digitized by Kansas Correctional Industries

Thesis (M.S.)--Wildlife and Fisheries Sciences Dept., South Dakota State University, 2009. / Includes bibliographical references (leaves 80-102). Also available via the World Wide Web.

In 1923 Svehla's government adopted regulation on the interim management of alcohol, which entrusted the management of distilling industry to private held company. The company, whose shareholders were organizations uniting most alcohol producers, as a result, found itself in the position of alcohol monopoly. In my thesis I show, that this unprecedented transfer of state reserves of alcohol and also all control over distilling industry into the hands of distillers was accompanied by widespread corruption that apply distilling representatives to lobby the leaders of the political parties in the relevant ministries and parliament.

Thesis (MSc Food Sc)--Stellenbosch University, 2000. / ENGLISH ABSTRACT: The distillery industry produces large volumes of waste water with a high organic content throughout the year. These effluents must be treated in some manner before being discharged or recycled in the factory. Several treatment options are in use presently, but they all have disadvantages of some nature, such as long retention times, bad odours or the need for large areas of land. Considerable interest has been shown in the application of anaerobic digestion, especially the UASB design (upflow anaerobic sludge blanket), to treat this high strength waste water. Thus, the aim of this study was to investigate the efficiency of an upflow anaerobic sludge blanket (UASB) bioreactor using full-strength distillery effluent. The activity of the bacteria in the biogranules was also evaluated by developing an easy and reliable activity method to estimate the general biogas and methanogenic activity and to calibrate this method using different anaerobic granules from different sources. The influence of high strength distillery effluent on the anaerobic digestion process was investigated using a mesophilic lab-scale UASB bioreactor. During the experimental study, the organic loading rate (OLR) was gradually increased from 2.01 to 30.00 kgCOD.m-3.d-1, and simultaneously, the substrate pH was gradually lowered from 7.0 to 4.7. It was found that at an OLR of 30.00 kgCOD. m-3.d-1,the pH, alkalinity and biogas production stabilised to average values of 7.8, 6 000 mg.l-1 and 18.5 I.d-1 respectively. An average COD removal> 90% was found indicating excellent bioreactor stability. The low substrate pH holds considerable implications in terms of operational costs, as neutralisation of the biorector substrate is no longer necessary. The accumulation of fine solids present in the distillery substrate was found at the higher OLR's and resulted in the granular bed increasing with subsequent biomass washout and a lowering in efficiency parameters. However, a possible pre-treatment filtration of these fine solids would eliminate this problem. The success of the upflow anaerobic sludge bed (UASB) process is mainly due to the capability of retaining the active biomass in the reactor. Over the years, several methods have been developed to characterise and quantify sludge activity but each has advantages and disadvantages. There is thus an increasing need for a rapid method to evaluate the activity of the granular biomass. The activity method of Owen et al. (1979) as adapted by Lamb (1995), was thus evaluated in terms of efficiency and applicability in determining the activity of granular samples. The method was found to be inaccurate as well as time consuming and it was thus modified. Results obtained with the modified assay method were found to be more accurate and the impact of the different test substrates (glucose, lactate, acetate and formate) on activity, was more evident. The activity of seven different anaerobic granules, was subsequently evaluated. Biogas (Ss) and methanogenic (SM) activity was not measured in volume of gas produced per unit COD converted or volatile suspended solids (VSS), but as tempo of gas production (ml.h-1) in a standardised basic growth medium. The activity data obtained were also displayed as bar charts and "calibration scales". This illustrative depiction of activity data gave valuable information about population dynamics as well as possible substrate inhibition. The "calibration scales" can also be used to group the general biogas (Ss) and methanogenic activities (SM) of any new biogranule relative to active (O-type) and inactive (W-type) anaerobic granules, providing that the same method of activity testing is used. The "calibration scales" can thus be used to give a fast indication of how the activity value of one sample relates to the activity values of other granules, even when using different test substrates. / AFRIKAANSE OPSOMMING: Die stokery industrie produseer groot hoeveelhede afvalwater, wat hoë ladings van organiese materiaal gedurede die hele jaar bevat. Hierdie afvalwater moet op een of ander manier behandel word voordat dit gestort of vir hergebruik aangewend kan word. Daar is tans verskeie behandelingsmetodes wat gebruik kan word, maar elk het sy eie tekortkominge soos bv. lang retensie tye, onaangename reuke of die behoefte aan groot stukke oop grond. Groot belangstelling is getoon vir die gebruik van anaerobiese vertering, en meer spesifiek die "uflow anaerobic sludge blanket" UASB bioreaktor vir die behandeling van stokery uitvloeisels. Die doel van die studie was dus om die algehele effektiwiteit van 'n UASB bioreaktor, wat onverdunde stokery uitvloeisel behandel, te evalueer. Die methanogene- en algehele aktiwiteit van die bakterië in die biogranules was ook ge-evalueer deurdat 'n maklike en betroubare aktiwiteitsmetode omtwikkel is, waarna hierdie metode ook toegepas was op 'n reeks van verskillende tipe biogranules. Die invloed van volsterkte stokery uitvloeisel op die anaerobiese verteringsprosesse was ondersoek met die gebruik van 'n mesofiele laboratoriumskaal UASB bioreaktor. Gedurende die eksperimentele studie, was die organiese ladingstempo (OLT) verhoog van 2.01 na 30.00 kgCSB.m-3.d-1 (CSB = chemiese suurstof behoefte) met die gelyktydige verlaging in die pH van die bioreaktorsubstraat van 7.0 na 4.7. Dit was vasgestel dat met 'n OLT van 30.00 kgCSB.m-3.d-1, die pH, alkaliniteit en biogas geproduseer, gestabiliseer het na gemiddelde waardes van 7.8, 6000 mg.-1 en 18.5l.d-1 , respektiewelik, sowel as 'n gemiddelde CSB verwydering van> 90%. Al hierdie waardes dui uitstekende bioreaktor stabiliteit aan. Die lae bioreaktorsubstraat pH kan van groot waarde wees vir die industrie, aangesien neutralisering van die uitvloeisel nie meer nodig is nie en kan sodoende die operasionele koste van die proses verlaag. Die konsentrering van fyn opgeloste soliedes in die bioreaktor by hoë OLT's, kan egter problematies raak, aangesien dit die granule-bed kan vergroot en veroorsaak dat van die biomassa uitspoel en kan verlore gaan. Die verlies van aktiewe biomassa kan die effektiwiteitsparameters negatief beinvloed, maar die plasing van 'n filterings stap voor die verterings stap, behoort hierdie probleem op te los. The sukses van die UASB-stelsel rus op die versekering dat die aktiewe biomassa in die reaktor behoue bly. Oor die jare was daar 'n verskeidenheid van aktiwiteitstoetsings-metodes ontwikkel, elk met sy eie nadele. Daar bestaan dus nog steeds 'n groot behoefte vir die daarstelling van 'n aktiwiteitstoetsings-metode wat vinnig en maklik is om uittevoer. Die aktiwiteitstoetsings-metode van Owen et al. (1979) wat deur Lamb (1995) aangepas is, was in terme van sy effektiwiteit en toepaslikheid ten opsigte van die gebruik daarvan vir aktiwiteitstoetsing vir biogranules, ge-evalueer. Dit is bevind dat die metode onakkuraat sowel as tydsrowend was en gevolglik dus aangepas. Die aangepaste metode het meer akkurate resultate gelewer en die impak van die verskillende toetssubstrate (glukose, laktaat, asetaat en formaat) op die granules het ook meer duidelik na vore gekom. Gevolglik was die aktiwiteit van sewe verskillende anaerobiese biogranules ondersoek. Die eenheid waarin atiwiteitsresultate aangegee is, was nie in volume gas geproduseer per eenheid CSB verwyder of per hoeveelheid gesuspendeerde vlugtige vetsure in die biomassa nie, maar as tempo van biogas (S8)- of metaan (SM)produksie (ml.h-1). Die data wat op hierdie wyse bekom was, is gebruik om staafdiagramme sowel as "kalibrasie skale" daar te stel. Hierdie illustrerende wyse om aktiwiteitsdata uit te beeld verskaf waardevolle informasie ten opsigte van die interaksies tussen die verskillende populasies in die granule en kan ook die aanwesigheid van moontlike substraat inhibisie aandui. Die "Kalibrasie skale" kan ook gebruik word om die algehele (SB) en methanogene (SM)aktiwiteite van einge nuwe biogranule vinnig te klassifiseer ten op sigte van 'n aktiewe (O-tipe) en 'n minder aktiewe (W-tipe) anaerobiese granules, mits dieselfde metode gebruik word om die aktiwiteits data te bekom.

Thesis (MSc Food Sc )--Stellenbosch University, 2003. / ENGLISH ABSTRACT: This study shows the potential of near infrared spectroscopy for both qualitative and quantitative analyses of wine and brandy in the distilling industry. Wines intended for further processing by the distillation industry have to meet certain specifications to ensure an efficient and cost-effective distilling process with a suitable product yield. Certain compounds have, therefore, been recognised as significant quality parameters and through their regulation, a product of consistent quality can be ensured. These quality parameters include alcohol, total acid, volatile acid, total sulphur dioxide, total phenolics, reducing sugars and acetaldehyde concentrations and pH. The effective control of all these compounds in the wines prior to distillation, is critical to maintain the feasibility and efficiency of the process. Fourier transform near infrared spectroscopy (FT-NIRS) can be used as a rapid and accurate analytical method for the determination of alcohol concentrations in distilling and brandy base wine. The choice of path length for the liquid sampling cell is an important consideration. In this study, alcohol with its strong absorption of OH bands in the near infrared region was not significantly affected, but difference in path length showed a significant influence on the prediction of the volatile acid concentration in distilling wine. Very strong correlations were found between the spectral data and the alcohol concentration data in distilling wine (r = 0.99, SEP = 0.18% v/v, RER =30) and brandy base wine (r = 0.92, SEP = 0.18% v/v, RER = 11.1). Good predictions were obtained for the volatile acid content of brandy base wine (r = 0.85, SEP = 0.04 g.L-1 ) but better accuracy could be obtained by incorporating a wider concentration range and more sensitive, although not the accepted reference methods, such as HPLC and GC, into the modelling. In contrast to the brandy base wine, the much cruder, turbid distilling wine (r = 0.67, SEP = 0.33 g.L-1 ) yielded poor calibration results for volatile acid. FT-NIRS can also be used as a rapid screening method to measure the total acid content, pH and total phenol levels in brandy base wine samples. Satisfactory predictions were obtained for the total acid content (r = 0.89, SECV = 0.38 g.L-1 ), pH (r = 0.84, SEP = 0.09) and total phenol levels (r = 0.71, SEP = 16.4 mq.L" GAE). The SEP of all the parameters compared well with the SEL and were within acceptable limits. The range of the concentration measurements were, however, very narrow and the accuracy of these models (measured as the ratio of the range to the standard error of prediction) were all below 10, indicating that modelling of these parameters in brandy base wine should be attempted with reference values covering a wider range. Poor correlations and predictions were obtained between the wine spectral data and the acetaldehyde (r = 0.39 , SECV =1.45 rnq.L"), residual sugar (r = 0.58, SECV =0.49 mq.L" ) and total sulphur dioxide (r = 0.35, SEP = 10.9 mq.L" ) concentration data. The unique character of potstill distillates, originating from the grape cultivar, type and age of the oak casks, lends itself to the development and blending of different brandies. The process of blending and correcting the brandy is necessary to obtain a product and brand of established and recognised character. The brandy is generally divided into different classes prior to blending based on the degree of "smoothness" or "hardness" of the matured distillate as determined by a sensorial panel. The ideal would, however, be to replace this subjective evaluation method with a more objective approach, such as an instrumental method like NIRS. Near infrared spectroscopy showed potential to successfully discriminate and classify brandy samples in terms of their sensorial classification status. The near infrared spectra of unblended three-year old brandies were measured and discriminant analysis performed using the subjective sensory classification data of the brandy. SIMCA (soft independent modelling of class analogy) models were developed by calculating separate peA (principal component analysis) models for each of the classes for which qualitative analysis were required. An assessment of class membership of an unknown sample was made on the basis of the distance to the cluster centroid. Two types of NIRS instruments, a Fourier transform near infrared system and a diode array spectrophotometer were employed to examine the discriminatory abilities of NIRS. Excellent predictions (90.9%-100%) were obtained for the "smoothest" and "hardest" classes within the 2000 and 2001 seasons with FT-NIRS. For the diode array spectra reasonable classification data were obtained for all the classes within the 2000 and 2001 seasons, as well as an excellent classification rate (100%) of the smoothest class in the 1999 season. Prediction results for class membership of the brandy samples from the three season's combined data did not deliver satisfactory discriminatory results. By applying NIRS and SIMCA, rapid and objective classification of brandy samples within seasonal context can be established to enhance the blending processes. Having accurate reference data in the calibration set is essential in near infrared spectroscopy (NIRS). The precision (usually decided by blind duplicate determination) of the reference data serves as the performance target for correlation-based near infrared analysis. The differences between reference values obtained from two independent laboratories and the Foss Winescan FT 120, which is also a predictive method, for the alcohol and volatile acid determinations in distilling wine and between two independent laboratories for the measurement of the total sulphur dioxide content, were investigated. ANOVA and post-hoc analysis (Bonferroni testing) revealed statistically significant differences (p<O.OS)between all three measurements for the volatile acid analysis. One laboratory differed significantly from the other two measurements in its measurement of the alcohol content in distilling wine. A paired t-test performed on the two laboratories' values obtained for total sulphur dioxide revealed statistically significant differences (p<O.OS) between the two sets of results. This study restated the importance of obtaining reliable reference data for calibration purposes to maintain the efficiency and accuracy and therefore the reputability of NIRS as an accurate and reliable analytical method. / AFRIKAANSE OPSOMMING: Hierdie studie dui op die potensiaal van naby infrarooi spektroskopie (NIRS) vir beide die kwalitatiewe en kwantitatiewe analise van wyn en brandewyn in die distillasie industrie. Sekere spesifikasies word gestel waaraan wyn met voldoen voor dit verdere prosessering tydens distillasie kan ondergaan. Die regulering van wynkwaliteit is noodsaaklik om die distilleringsproses ekonomies volhoubaar te maak. Sekere komponente in die wyn het 'n direkte invloed op die kwaliteit van die eindproduk en word beskou as betekenisvolle parameters wat beheer moet word om 'n produk van konstante hoë kwaliteit te lewer. Hierdie parameters sluit alkohol, titreerbare suur- en vlugtige suurinhoud, asook die pH, totale swaweidioksied, totale fenol vlakke, reduserende suikers en asetaldehied konsentrasies in. Die regulering van al hierdie komponente in die wyne is essensieel voor die aanvang van distillasie om die effektiwiteit en produktiwiteit van die proses te verseker. Fourier transformasie naby-infrarooi spektroskopie (FT-NIRS) kan benut word as 'n vinnige en akkurate analitiese metode om die alkohol konsentrasie in stook- en rabatwyn te bepaal. 'n Geskikte padlengte kuvet moet gekies word vir die ontleding van vloeistof monsters met NIRS. In hierdie studie is gevind dat die voorspelling van alkoholkonsentrasie in wyn, nie soseer beïnvloed is deur die keuse van padlengte nie. 'n Vlugtige komponent soos asynsuur, wat die maatstaf van die vlugtige suurinhoud van die wyn is, word egter regstreeks beïnvloed deur die keuse van padlengte. 'n Sterk korrelasie is gevind in die datastel vir die alkoholinhoud van die stookwyn (r = 0.99, SEP = 0.18% v/v, RER = 30) en rabatwyn (r = 0.92, SECV = 0.18% v/v, RER = 11.1). Baie goeie voorspellings is ook gevind vir die vlugtige suurinhoud van rabatwyn (r = 0.85, SEP = 0.04 g.L-1). Die akkuraatheid van die metode (uitgedruk as die RER waarde) het egter aangedui dat 'n wyer konsentrasie omvang en meer akkurate verwysingswaardes gebruik moet word om 'n geskikte model vir vlugtige suur kwantifikasie in rabatwyn te bewerkstellig. In teenstelling met die resultate verkry vir die rabatwyn, het die ruwer, turbiede stookwyn swak gekalibreer vir die vlugtige suurinhoud (r = 0.67, SEP = 0.33 g.L-1 ). FT-NIRS kan ook gebruik word as 'n vinnige seleksie metode om titreerbare suur vlakke, pH en totale fenolinhoud van rabatwyn monsters te bepaal. Sterk korrelasies is in die datastel gevind vir die titreerbare suurinhoud (r = 0.89, SECV = 0.38 g.L-1 ), pH (r = 0.84, SEP = 0.09) en totale fenol vlakke (r = 0.71, SEP = 16.4 rnq.L" GAE) van rabatwyn. Die SEP waardes verkry vanaf die NIRS bepalings het baie goed vergelyk met die standaard foute wat verkry is vir die chemiese bepalings. Die konsentrasie omvang van die verwysingswaardes was, egter, beperk en die akkuraatheid van die NIRS modelle (uitgedruk as die verhouding tussen die konsentrasie omvang en die standaard fout van voorspelling) was laer as 10. Dit het daarop gedui dat die NIRS kalibrasie van die parameters eerder uitgevoer moet word met monsters wat 'n wyer konsentrasie omvang dek en gemeet is met meer sensitiewe analitiese metodes soos HPLC en GC. Die kalibrasie en voorspelling van die asetaldehied inhoud (r = 0.39 , SECV =1.45 rnq.L"), reduserende suiker konsentrasie (r = 0.58, SECV =0.49 mq.L" ) en totale swaweldioksiedinhoud (r = 0.35, SEP = 10.9 rnq.L" ) het onaanvaarbare voorspellings gelewer. Die unieke karakter van potketel distillate wat bepaal word deur die druifkultivar, tipe en ouderdom van die eikehoutvate, bied die potensiaal vir die ontwikkeling en versnyding van verskillende style brandewyn. In 'n poging om 'n produk en handelsnaam van erkende en gevestigde karakter te verkry, moet die brandewyn eers versny word na veroudering. Die brandewyn word geklassifiseer in verskillende style op grond van sensoriese gehalte. NIRS is gebruik as 'n vinnige tegniek om te onderskei tussen verskillende style onversnyde drie jaar-oue brandewyn. SIMCA (soft independent modelling of class analogy) modelle is ontwerp deur aparte PCA (hoof komponent analise) modelle vir elke klas waarvoor kwalitatiewe analise vereis is, te ontwikkel. 'n Skating van klas lidmaatskap van 'n onbekende monster is gemaak deur die afstand van die monster na die groepsmiddelpunt te bereken. Twee soorte NIRS instrumente, 'n FT-NIR en 'n diode array spektrofotometer is gebruik tydens die studie om die diskrimineringsvermoë van elk te ondersoek. Uitstekende voorspellings (90.9%-100%) is gekry vir die "sagste" en "hardste" klasse binne die 2000 en 2001 seisoene met FTStellenbosch NIRS. Met die diode array spektra is redelike klassifikasie resultate verkry vir al die klasse binne die 2000 en 2001 seisoene, asook 'n uitstekende klassifikasie (100%) vir die sagste klas binne die 1999 seisoen. Voorspellings vir klas lidmaatskap van die brandewyn monsters vir die drie seisoene se gekombineerde data het egter onaanvaarbare diskriminasie resultate gelewer. NIRS en SIMCA sal 'n vinnige en objektiewe klassifikasie proses van onversnyde brandewyn monsters kan meebring wat tot voordeel van die versnydings- en produksieproses sal wees. Hierdie voorspellings sal egter binne seisoensverband moet geskied. Akkurate verwysingsdata in die kalibrasie datastel is 'n essensiële fasset van naby infrarooi spektroskopie. Die presisie (bepaal deur blinde duplikaat bepalngs) van die verwysingsdata dien as 'n prestasie teiken vir korrelasie-gebaseerde naby infrarooi voorspellings. Die verskille tussen die data verkry vir die alkohol- en vlugtige suurinhoud van stookwyn deur twee onafhanklike laboratoriums en 'n instrumentele metode (Foss Winescan FT 120) asook die verskille tussen die data verkry van twee laboratoriums vir die totale swaweidioksied metings, is ondersoek. ANOVA en post-hoc toetse het aangedui dat al die data verkry vir die vlugtige suur bepalings, statisties betekenisvol van mekaar verskil het (p<0.05). Vir die alkohol bepalings in stookwyn, het die data verkry van een laboratorium betekensvol verskil van die ander twee datastelle. Gepaarde t-toetse uitgevoer op die data verkry vir totale swaweldioksiedinhoud, het ook betekinisvolle verskille aangedui tussen die twee laboratoriums. Hierdie studie het die belangrikheid van betroubare verwysingsdata vir kalibrasie doeleindes beklemtoon, veral waar 'n tegniek soos NIRS se werksverrigting afhang van die akkuraatheid van dié waardes.

Submitted by Filipe dos Santos (fsantos@pucsp.br) on 2018-07-25T11:56:20Z No. of bitstreams: 1 Priscila Santesi Bianchini Amore.pdf: 1833291 bytes, checksum: 979fdd6ac345f6d9f95fa999697287aa (MD5) / Made available in DSpace on 2018-07-25T11:56:20Z (GMT). No. of bitstreams: 1 Priscila Santesi Bianchini Amore.pdf: 1833291 bytes, checksum: 979fdd6ac345f6d9f95fa999697287aa (MD5) Previous issue date: 2018-03-15 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / This work has an objective to analyze the appearing of petroleum industry on XIX century, from several researches which had been realized to obtain new sources of combustibles by distillation of mineral coal, vegetal and petroleum, including for illumination. They tried to analyze specifically the research contributions of Abraham Gesner (1846 – 1854) related to kerosene production, on a period which were proliferated researches related to development of new kind of combustibles. The development of this study was guided mainly on the analyze of the published work by Gesner, A Practical Treatise on Coal, Petroleum and Other Distilled Oils, where this publication became the main reference job for installation of petroleum distillery on second half of XIX century, on Unites States. It is possible to evidence with the publication of his book, Gesner, on the same time he had as a goal to make this job a practical manual for the installation and production of oils from the distillation of coal, bitume and petroleum, it also aimed to provide answers to his opponents on the several judicial disputes related to patents where he was involved / O presente trabalho tem por objetivo analisar o surgimento da indústria do petróleo no século XIX, a partir das várias pesquisas que estavam sendo realizadas a fim de se obter novas fontes de combustíveis por destilação de carvão mineral, vegetal e petróleo, inclusive para iluminação. Procurou-se analisar especificamente as contribuições das pesquisas de Abraham Gesner (1846-1854) relacionadas à produção do querosene, num período em que proliferavam as pesquisas relacionadas ao desenvolvimento de novos combustíveis. O desenvolvimento deste estudo pautou-se principalmente na análise da obra publicada por Gesner, A Practical Treatise on Coal, Petroleum and Other Distilled Oils, publicação que se tornou a principal obra de referência para a instalação das destilarias de petróleo na segunda metade do século XIX, nos Estados Unidos. Pode-se evidenciar que, com a publicação de seu livro, Gesner, ao mesmo tempo que tinha por objetivo fazer desta obra um manual prático para a instalação e produção de óleos a partir da destilação de carvões, betume e petróleo, também visava apresentar respostas aos seus adversários nas várias disputas judiciais relativas às patentes nas quais se envolveu

In the last decade, the machine learning universe embraced deep neural networks (DNNs) wholeheartedly with the advent of neural architectures such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), transformers, etc. These models have empowered many applications, such as ChatGPT, Imagen, etc., and have achieved state-of-the-art (SOTA) performance on many vision, speech, and language modeling tasks. However, SOTA performance comes with various issues, such as large model size, compute-intensive training, increased inference latency, higher working memory, etc. This thesis aims at improving the resource efficiency of neural architectures, i.e., significantly reducing the computational, storage, and energy consumption of a DNN without any significant loss in performance. Towards this goal, we explore novel neural architectures as well as training algorithms that allow low-capacity models to achieve near SOTA performance. We divide this thesis into two dimensions: \textit{Efficient Low Complexity Models}, and \textit{Input Hardness Adaptive Models}. Along the first dimension, i.e., \textit{Efficient Low Complexity Models}, we improve DNN performance by addressing instabilities in the existing architectures and training methods. We propose novel neural architectures inspired by ordinary differential equations (ODEs) to reinforce input signals and attend to salient feature regions. In addition, we show that carefully designed training schemes improve the performance of existing neural networks. We divide this exploration into two parts: \textsc{(a) Efficient Low Complexity RNNs.} We improve RNN resource efficiency by addressing poor gradients, noise amplifications, and BPTT training issues. First, we improve RNNs by solving ODEs that eliminate vanishing and exploding gradients during the training. To do so, we present Incremental Recurrent Neural Networks (iRNNs) that keep track of increments in the equilibrium surface. Next, we propose Time Adaptive RNNs that mitigate the noise propagation issue in RNNs by modulating the time constants in the ODE-based transition function. We empirically demonstrate the superiority of ODE-based neural architectures over existing RNNs. Finally, we propose Forward Propagation Through Time (FPTT) algorithm for training RNNs. We show that FPTT yields significant gains compared to the more conventional Backward Propagation Through Time (BPTT) scheme. \textsc{(b) Efficient Low Complexity CNNs.} Next, we improve CNN architectures by reducing their resource usage. They require greater depth to generate high-level features, resulting in computationally expensive models. We design a novel residual block, the Global layer, that constrains the input and output features by approximately solving partial differential equations (PDEs). It yields better receptive fields than traditional convolutional blocks and thus results in shallower networks. Further, we reduce the model footprint by enforcing a novel inductive bias that formulates the output of a residual block as a spatial interpolation between high-compute anchor pixels and low-compute cheaper pixels. This results in spatially interpolated convolutional blocks (SI-CNNs) that have better compute and performance trade-offs. Finally, we propose an algorithm that enforces various distributional constraints during training in order to achieve better generalization. We refer to this scheme as distributionally constrained learning (DCL). In the second dimension, i.e., \textit{Input Hardness Adaptive Models}, we introduce the notion of the hardness of any input relative to any architecture. In the first dimension, a neural network allocates the same resources, such as compute, storage, and working memory, for all the inputs. It inherently assumes that all examples are equally hard for a model. In this dimension, we challenge this assumption using input hardness as our reasoning that some inputs are relatively easy for a network to predict compared to others. Input hardness enables us to create selective classifiers wherein a low-capacity network handles simple inputs while abstaining from a prediction on the complex inputs. Next, we create hybrid models that route the hard inputs from the low-capacity abstaining network to a high-capacity expert model. We design various architectures that adhere to this hybrid inference style. Further, input hardness enables us to selectively distill the knowledge of a high-capacity model into a low-capacity model by cleverly discarding hard inputs during the distillation procedure. Finally, we conclude this thesis by sketching out various interesting future research directions that emerge as an extension of different ideas explored in this work.